WO2016058560A1 - External acceleration method based on serving end and external buffer system for computing device, and device implementing said method - Google Patents

Abstract

An external method based on a serving end and an external buffer system for a computing device, and a device implementing said method, comprising: at least a serving end buffer device (S) connected via a high-speed or multi-channel data transmission device (L) to an accelerated computing device (C). L may be of a wired transmission type or a wireless transmission type. S provides C with a relatively high-speed write buffer (R layer, such as RamDisk) and a random read buffer layer (D layer, such as a phase change memory), and buffers files frequently used by the system and programs and frequently read and written files for C, transfers I/Os directed at the C hard drive, and provide computational support to C via a virtual layer (V). When L is of a wired type, L may be an optical fiber, a transmission line above USB 3.0, a CAT5E network cable, or other high speed wired transmission devices. When L is of a wireless type, L may also be a wireless network card supporting the WiGig technology, or a C base or an external device connected to S by means of wireless transmission.

Description

External computing device acceleration method based on server and external cache system and device implementing the same Technical field

The product belongs to the field of computer equipment, and is an electronic device acceleration method based on cross-device cache technology, and a device for implementing the method.

Background technique

At present, both personal users and enterprise users use computers and mobile devices in large quantities, and in the future, robots and smart home devices and smart wearable devices may be widely used. However, the update of these electronic smart devices is very fast, and the product models are numerous, the types of devices are numerous, the age span is large, the system platform is complex, and the performance upgrade is extremely difficult. Especially for general mobile devices, embedded smart devices and wearable devices, there is basically no way to upgrade.

The problem is currently lacking an effective universal upgrade solution.

1. Why do you need to accelerate the upgraded product?

The development of technology always puts the hardware behind. Both software and systems are growing rapidly. Upgrading computers costs a lot, usually thousands of dollars, while mobile phones, tablets, etc. can hardly be upgraded. At present, this kind of upgrade is a thorny problem. The existing solution is usually equipped with a new machine. No matter the computer or the iphone, it costs nearly 10,000, and it is very troublesome. What's worse, the old machine has become e-waste.

Of course, there are also many users who buy spare parts to disassemble and replace parts, commonly known as DIY market, but the technical requirements are higher and the difficulty is also great. For example, changing the CPU, changing the hard disk not only needs to accurately connect various data lines in the chassis and The socket also needs to export the data of the old hard disk and reinstall the system and various types of software. The average user will not. And the cost is still high, and the compatibility with the motherboard is also a big problem.

This is still for the traditional computer, for mobile computing devices, high-skilled users who are good at DIY are also unable to do anything.

There are also some software that can optimize the system, such as the 360 optimization master, accelerate the ball, but these do not substantially improve the hardware capabilities, just clean up the garbage, etc., and do not enhance the performance of the computer itself.

2. What are the bottlenecks in the performance of computers and mobile devices?

It is the speed of storage devices such as hard disks, especially the frequent reading and writing of small files and random reading and writing.

Over the past decade, CPU and memory performance has increased more than 100 times, but the performance of hard drives has only doubled. The bottleneck of the entire data processing is on the hard disk. As long as this bottleneck can be opened, the information transmission will be on the "highway." (The situation is similar for mobile devices, where the processor performance of mobile devices is growing rapidly, but the performance of memory chips is growing. slow. )

Mobile devices are small in size and generally do not provide a DIY upgrade. The following describes the situation on the computer.

On your computer, a solid state drive can be used for the upgrade. SSDs do not have the motors and rotating media of ordinary hard drives, so they are quick to start and have excellent shock resistance. Solid state drives do not use heads, disk reads and writes are fast, and latency is small. The read and write speed can generally reach more than 100M per second. Although the speed is much faster than the mechanical hard disk, but the disadvantages are also many, such as expensive, small capacity, short battery range, limited write life and so on. Often not expensive, the capacity is small, and the entry-level Kingston SSD NOW of about 500 yuan has only 32G capacity. The large capacity is expensive, the same is 1TB size, the mechanical hard disk is about 200 yuan, and the solid state hard disk is at least 5,000. Therefore, in the new factory computer, the solid state drive still does not replace the mechanical hard disk.

And the old equipment upgrade should mainly consider the feasibility, as well as the cost performance, these two aspects. Feasibility: The first is the compatibility issue. Early motherboards did not support solid state drives. Specifically, first of all, many computers from early to before 2009 did not support SSDs, and computers that were 11 years ago did not support SATA 2 protocol at all. The maximum support speed of the motherboard interface is 100 MB per second for ordinary IDE or 150 MB per second. The SATA hard disk protocol cannot be accelerated with SSD at all. It is almost impossible to replace the motherboard. Secondly, it is still inconvenient. The average user is not good at replacing the hard disk. Changing the hard disk means changing the whole system, copying all the files, reinstalling various drivers, and consuming at least one or two days. Moreover, the SSD setting is complicated, and it can only exceed the normal hard disk speed under Win7 or Win8. XP does not recognize the Trim command of SSD, 4k alignment and ACHI. Not only can't you speed up, you can't use your computer, and in most cases it will blue screen and crash.

In terms of cost performance: the first is the high price, the price of the entry-level 32G-64G for SSDs is about 500 yuan, but 64G has basically no space after installing Win7 system and Office. The price of the entry-level 128G has already approached 1,000 yuan. The cost of upgrading an old computer is not worth it. The second is short life, solid state drives are generally MLC flash memory, and its life is very short without proper maintenance. Turning on Trim and setting 4k alignment and other maintenance measures will not be available to customers.

3. So is there any other low-cost and more convenient technical solution to solve the bottleneck of hard disk speed?

The mobile device is small in size and generally does not provide a DIY upgrade method, and an upgrade method is urgently needed.

The following describes other upgrade scenarios on your computer.

Other devices are also available on the computer to speed up the attempt. Currently known is Intel's Fast Disk: Fast Disk is a PCI-E interface expansion card, equipped with one or two MLC NAND flash memory, as a Mini PCI-E 1x specification expansion card, through the PCI-E bus Data exchange with the system I/O controller. The flash memory module used by the Fast Disk is NAND, not NOR. This is because NAND is better than NOR in accessing data performance and has better performance. Value for money.

With the support of the system, ReadyBoost and ReadyDrive functions are provided. These functions will directly improve the performance of the system in terms of booting, hibernation, installing programs, copying files, loading games and other tasks related to disk operations. According to official data, the fast disk can speed up the boot speed by 20%, while reducing the number of hard disk revolutions to save power.

Introduction to ReadyBoost:

When ReadyBoost determines that the cache in the flash is more capable of satisfying random reads than the cache in the hard disk, it will randomly read the data from the flash disk media. The hard disk will read a large amount of data in batches at a time, and temporarily store it in the fast disk for the system to call at any time; at the same time, the data to be written is temporarily stored in the fast disk, and then accumulated to a certain number and then uniformly written to the hard disk. This on-demand read/write mechanism is very helpful in improving system performance. During this time, the hard disk is idle, and the larger the capacity of the fast disk, the longer the idle time of the hard disk, thereby reducing the number of mechanical rotations and power consumption, and prolonging the battery life of the notebook battery.

Introduction to ReadyDriver:

ReadyDrive is in fact Microsoft's name for a hybrid hard drive (a hard drive with internal flash components). In addition to the obvious random access speed advantage of flash memory, the biggest temptation is that the data stored in it is "right to wait" - because for flash memory, there is no need to start the head or wait for the head to rotate to the proper position. Hybird drives start up, hibernate, sleep faster, and consume less power. Because when the operating system reads and writes the cache, the drive itself can temporarily stop working without consuming any power. When resumed from hibernation, the laptop can immediately start reading data from the cache, instead of waiting for the drive's head to start up as usual.

In the driver of the fast disk, it can be seen that the user can set the module to provide ReadyBoost, ReadyDrive function through the software interface, or both.

However, the fast disk is still not an effective upgrade program. It is precisely because of this that it has not been mentioned. The main reasons for its failure are: 1. It can't be used on desktop computers, nor can it be used in most notebooks. All Netbooks and most laptops do not support the Fast Disk module, because this not only requires the laptop to provide an additional Mini PCI-E slot, but more importantly, the laptop's SATA interface supports the ACHI function; 2. The installation is complicated. The average user does not disassemble the mini PCI-E, so that it cannot be used for old computer upgrades; 3. The effect is not good. The speed of the PCI-E bus itself is limited to 150M per second, and Intel's flash memory is far from this speed. The measured random read and write speed of 35M per second is not much improved for the hard disk, and is not as good as the solid state hard disk. Intel's fast disk is limited in size, unable to add cache or parallel modules, or more master ICs; 4. expensive. 4G's fast disk pricing is at $100; 5. System compatibility is poor. This alone is enough to rule out the possibility that the Turbo is used to speed up old computers. Both Readydrive and Readyboost can only be used for Windows Vista and above, and most of the old computers are XP operations. The system can only run smoothly under XP.

In summary, large-volume devices such as PCs and notebooks in smart devices have complex and difficult upgrade problems, while small-sized devices such as mobile devices have problems that cannot be upgraded.

Summary of the invention

In view of the above problems, the present invention provides an external computing device acceleration method based on a server and an external cache system and an apparatus for implementing the method.

The method achieves acceleration by composing a performance-classified, cross-device caching system, including at least one server-side caching device (hereinafter referred to as S), through a high-speed or multi-channel data transmission device (hereinafter referred to as L) and an accelerated computing device (hereinafter referred to as C) connection. L can be either a wired transmission type or a wireless transmission type. S provides a relatively high-speed write cache layer (such as Ramdisk, etc.) and a random read cache layer (such as resistive memory) for C, a common file for C cache system and programs, and frequently read and write files, etc., which are transferred to the C hard disk. I/O and provide computational performance support for C through the virtual layer. When L is a wired type, L may be an optical fiber, a USB 3.0 or higher transmission line, a super five type network cable, or other high speed wired transmission equipment. When L is a wireless type, L may be a wireless network card supporting WiGig technology, or a base or an external box of C that is connected to the S end by wireless transmission. See Figure 1.

The computing device here does not specifically refer to a computer, but should include a variety of computers, tablets, mobile phones, robots or smart hardware in a broad sense. These computing devices contain data and file acquisition and processing. These generalized computing devices can be used to form a short-distance, multi-channel, cross-device caching system for the same enterprise or the same family.

S accelerates C through the three acceleration layers of the division of labor described above, typically as the write cache layer (hereinafter referred to as R Layer or R layer, such as Ramdisk, etc., S simulates memory as a disk and creates a cache for C in it). To obtain a larger cache speed), random read cache layer (hereinafter referred to as D Layer or D layer, such as resistive memory, or NAND disk array combined in RAID mode), virtual layer (hereinafter referred to as V Layer or V layer) Provide computational performance support for C by virtualizing application or hardware. Of course, due to the consideration of power consumption and cost in a specific environment, the above three layers can be strengthened, merged or reduced, and should be determined according to specific needs. For example, for an enterprise application environment, the V layer may need multiple types to suit different application requirements. For a simple home application, the R layer and the D layer may be physically combined into one RD layer; for the C terminal, idle availability is available. In the memory device, S can retrieve the memory of the C-end device part and the cache of the S device to form a complex cache across devices.

In addition to establishing a cache system, S can also provide computational performance support for C. The method can be any one or more of the following three methods: 1. S-shared computing tasks; 2. Through virtual architecture in S, allow C Run this virtual layer in S in a remote operation mode, display the interface on C and realize interaction with the user. One S can create multiple Cs. The virtual architecture is built, and the S resources are deducted according to the real-time usage of C. The virtual architecture separated in S can be a virtual machine or a virtual application layer. 3. The S virtualizes the application to pre-store the program files required by C. The system environment required by the program is in the S device, and C can run the application directly in S.

It is conceivable that S and C will have a rich topological relationship: First, S and C can be one-to-one, one-to-many, or many-to-many relationships, such as a home environment may be one-to-one or one For many relationships, the environment of enterprises, schools, administrative units, etc. may be one-to-many or many-to-many relationships (this topological relationship is hereinafter referred to as elastic service relationship); second, S and C are only relative relationships, ie one The C in the group relationship can be S in another group relationship, such as an accelerated computer, which can be accelerated for another computer at the same time, and correspondingly, the S in a group relationship can also be in another group relationship. C, such as the S-acceleration of the next-level performance and cache speed by the computing device of the higher-level performance and cache speed (the topological relationship is hereinafter referred to as the performance flow relationship); three, multiple Cs can simultaneously act as S, forming each other Accelerated networking of P2P to improve performance (for example, a computer C1 has a large memory, can provide a stronger small file cache capability and write operation cache, but the disk speed is relatively insufficient, and a computer C2 has multiple blocks Large flash solid state hard Disk, and increase the bandwidth in the form of RAID, with a stronger read operation cache, then C2 can be S with C1, C2 and C1 write cache is more written to C1, and read operation cache is more written to C2 This topological relationship is hereinafter referred to as a performance complementarity relationship).

These rich topological relationships can extend more applications. For example, multiple Ss are networked in a high-speed connection such as fiber to enhance read-ahead analysis and achieve higher cache performance. As another example, from a higher-level S, a higher-speed L device, such as a multi-path fiber, delivers performance to the next-level S, thereby taking advantage of the idle computing center's idle performance.

When L is a wired type, L can be a fiber optic cable, a Thunderbolt transmission line, an extended USB 3.0 or higher transmission line, a super five network cable, or any other type of wired transmission device with a transmission speed greater than 60 MB per second, and S is directly connected through these transmission cables. To C. Since the current bandwidth utilization efficiency of the USB protocol is low, when the USB interface is connected to the USB interface, the USB protocol can be improved, the BOT protocol in the traditional USB interface protocol is optimized, and resource allocation is performed on the USB transmission protocol. optimization.

When L is a wired type, an interface conversion method can be adopted for a device that does not have a high-speed wired network interface. For example, a high-speed network cable and a USB Ethernet adapter with a CAT6 standard or higher (such adapters have a USB interface on one end and an Ethernet interface on the other end, and an Ethernet port is added to the computer through an existing USB input port) And other necessary accessories, use the USB end of the adapter to connect S or C, and use the high speed network cable above cat6 standard to transmit between the two.

L is a wireless type, L can be a wireless network card supporting WiGig technology (the technology can currently transmit data at a speed of 25 Gbits per second), or a wireless network card that can be an 802.11ac for a mobile device. Because the internal disk reading and writing of the mobile device itself is very slow, the bandwidth of 802.11ac can also be significantly improved.

When L is a wireless type, an interface conversion method can also be adopted for a device that does not have a high-speed wireless network interface. For example, L can be a base, an external box or a protective case that is transmitted between the wireless transmission and the server (the base is connected to the accelerated computer by a USB interface or a Thunderbolt interface or other interface), or any other type including wireless The part of the device is transmitted, and S is connected to C through wireless transmission via L.

This type of interface conversion method is more practical for mobile devices. For example, L is a base or a protective case that is transmitted between the wireless transmission and the server. The base can communicate with the S with a high-speed wireless network such as WiGig, and then with a USB interface (all USB interfaces mentioned in this document are generalized). USB interface, including various branches such as MicroUSB, MiniUSB, Wireless USB, etc.) or Thunderbolt interface or other interface to C, C does not need high-speed wireless communication capabilities, can also be a normal tablet or mobile phone, using the base When the case is protected, you can enter the high-performance mode and return to a lighter but lower-performance mode without using the base or case.

In order to achieve higher transmission performance, L can adopt a multi-channel method of increasing bandwidth. At this time, the connection between S and C passes more than one L, or L contains more than one interface connected with C, thereby increasing the transmission bandwidth (Example 1, C is connected to S through its own Wifi, on the one hand, The Wifi adapter of the USB interface is also connected to the S, and the bandwidth is merged after the connection; for example, the plurality of USB interfaces of the C are connected to the S by using the Wifi adapter of the USB interface, thereby increasing the upper limit of the Wifi bandwidth to the upper limit of the USB bandwidth; Wigig's base L communicates with S via Wigig and is connected to C by Bluetooth and MicroUSB.

This method of grading across devices will also change the traditional cache implementation. In the past, cache implementations were often self-learning based on the algorithm of a single device, and in the new system, the cache work itself became a source of big data. The advantage is that the cache structure can be optimized or pre-determined according to the statistics of various applications and related files cached by multiple C-sides served by a single or multiple S pairs (Example 1: a large number of C-sides A certain folder of a game program exhibits a frequent reading feature. When S services a new C-end, if the program is found, the pre-judgment work can be directly performed, such as caching the file frequently read and written on other devices. Clipping to high-speed devices without re-accumulating cached data; Example 2: A large number of programs on the C-end display frequent write jobs, such as a shopping browser, when the browser is launched, it can be prejudged Allocate a large S-side write cache layer without re-accumulating cached data). In fact, many programs cannot learn the optimal cache on a single device because the user is not used frequently. However, the acquisition of cross-device data enables statistics and judgment of a large number of data samples, making many rarely used programs. Even the programs used for the first time can be accurately pre-optimized.

Various systems can be designed or manufactured based on the above methods. Two typical devices are manufactured in the example of the present invention, one is a typical wired type system design for enterprises and the other, and the other is a wireless type system design for home and individual users, respectively, as shown in FIG. 2 and the accompanying drawings. As shown in Figure 3, these two devices will be introduced in the following scenario examples. Make a specific introduction. For home and personal user equipment, for the sake of mute and energy saving, the S in the second example can be in the standby state when it is not connected to C, and wakes up through the C connection.

These devices designed specifically for the methods described herein have better results, as well as lower power consumption and cost. Of course, in some cases, if there are original equipment for retrofitting, and without considering power consumption and cost, S can be virtualized on existing large computers, servers, or high-performance computers. R layer D layer V layer, etc., and install appropriate modifications to be converted into S devices, then install drivers and L server on S and C to form a cross-device cache system, and achieve the purpose of C acceleration. Methods and apparatus for retrofitting an acceleration system are also provided in the sample portion of the invention.

Advantageous effects of the present invention

Compared with the traditional computer upgrade, the method and the corresponding device for implementing the method have the following advantages:

1. Support a wide range of equipment types, which is an effective and feasible operation scheme: upgrading old computers often requires disassembling and replacing memory for hard disk. If you want to speed up, you need to change the motherboard to change the CPU, and often work for a day or two. Or a blue screen, the compatibility between the various interfaces is not clear to the average user. The most appropriate way is to use the porter to carry the computer to the computer city to upgrade the site, but the price is very high, the cat is a lot more tired, and often the parts are stolen. The method and the device implemented are very simple to complete. However, there is no effective solution for upgrading mobile devices in the past, and the present invention provides a feasible cross-device performance enhancement scheme.

2. Support batch equipment acceleration at the same time: For enterprises and schools with a large number of equipment, only a very low cost can be used to upgrade batch equipment.

3. Support short-range wireless acceleration: Internet of Things and mobile devices may be widely used in the future. These devices are often small in size, neither have an external interface nor a wired interface, and are also likely to be embedded devices and cannot be upgraded. Replacement equipment must be replaced in a complete set, which is costly. For the past, the mobile device itself provides simple performance and relies on the mobile Internet to obtain information flow. The present invention considers that long-distance networks acquire information and short-range networks acquire performance. The solution of the present invention provides another aspect, that is, by short-distance cache acceleration and performance transfer (the distance is much smaller than that of the network mentioned in the past), the mobile device, the wearable device, and the Internet of Things smart device can be accelerated. Mobile devices are limited by power consumption and volume. The processing performance and cache media are inherently insufficient, but mobile devices often have high wireless data speeds.

4. The effect is obvious: taking the sample of the present invention as an example, for a computer with a common mechanical hard disk of a common 100 Mbps network card, after the bandwidth is merged by using a USB modified Gigabit network card and a multi-channel network, the running speed of some programs can be improved by 1-3. Times, (In addition, in fact, for general computers, you can transfer USB3.0 from PCI-E or ExpressCard. Compared with the original USB3.0, these transferred USB3.0 speed is lower, data transmission is about At 150M per second. So the old computer can also use USB3.0), for the newer mechanical hard disk or hybrid hard disk computer of Gigabit LAN, the program starts running. The speed can be increased by 20%-80%. For the latest PCs using WIGIG technology SSDs, such as DELL LATITUDE 7440, theoretically (calculated at 25Gbits per second), the program startup speed can still be increased by 2-3 times.

5. Low cost: The cost of the program is within the tolerance of general business units and individual consumers.

6. Sustainable upgrades: Further upgrades to C can be achieved through further upgrades to S.

7. It can be upgraded step by step: further upgrade of S can be achieved by the same method with higher level S device acceleration.

8. Rely on big data intelligence to enhance cache performance: This hierarchical cross-device approach will also change the traditional cache implementation. In the past, cache implementations were often self-learning based on the algorithm of a single device, and in the new system, the cache work itself became a source of big data. The advantage is that the cache structure can be optimized or pre-determined according to the statistics of various applications and related files cached by multiple C-sides served by a single or multiple S pairs (Example 1: a large number of C-sides A certain folder of a game program exhibits a frequent reading feature. When S services a new C-end, if the program is found, the pre-judgment work can be directly performed, such as caching the file frequently read and written on other devices. Clipping to high-speed devices without re-accumulating cached data; Example 2: A large number of programs on the C-end display frequent write jobs, such as a shopping browser, when the browser is launched, it can be prejudged Allocate a large S-side write cache layer without re-accumulating cached data). In fact, many programs cannot learn the optimal cache on a single device because the user is not used frequently. However, the acquisition of cross-device data enables statistics and judgment of a large number of data samples, making many rarely used programs. Even the programs used for the first time can be accurately pre-optimized.

9. With topological scalability, it will help the future development: First, S and C can be one-to-one, one-to-many, or many-to-many relationships, such as the home environment may be one-to-one or one-to-many. Relationship, and the environment of enterprises, schools, administrative units, etc. may be one-to-many or many-to-many relationships (this topological relationship is hereinafter referred to as the elastic service principle); second, S and C are only relative relationships, that is, a group of relationships C can be another set of relationships S, such as an accelerated computer, can simultaneously accelerate for another computer, and correspondingly, S in a set of relationships can also be C in another set of relationships, such as S-acceleration of the next-level performance and cache speed by the computing device of the higher-level performance and cache speed (this topological relationship is hereinafter referred to as the performance fluid principle); three, multiple Cs can simultaneously act as S to form P2P acceleration Networking for performance improvement (for example, a computer C1 has a large memory, can provide a stronger small file cache capability and write operation cache, but the disk speed is relatively insufficient, and a computer C2 has a large number of large Flash SSD with RAI The D form increases the bandwidth and has a stronger read operation buffer. C2 can be S with C1, C2 and C1 write buffers are more written to C1, and read operations cache more writes to C2. Hereinafter referred to as the principle of performance complementarity).

Implementation case of the content of the present invention

The present invention has been tested successfully for four examples.

The first example is an active wired acceleration center for enterprises, schools, and schools. This example assumes the applicable ring. The environment is an enterprise and enterprise. The unit has 50 computers that have been used for many years and are all ordinary mechanical hard disks (sequential read and write speed is about 40-70MB per second, the key is that 4K random read and write is very slow at around 1MB per second) . The equipment of enterprises, institutions and schools is mainly desktop computers. The static environment does not need to consider mobile issues, and is more suitable for wired L. Commercial and government, some departments must be wired for security and confidentiality, and wireless network cards will be taken out.

The acceleration scheme connection diagram of the first example is shown in FIG.

Part S: Specialized cache service machine with silent design, 10 Gigabit Ethernet card, 4 Ethernet outlets and 50 Ethernet outlets through routers, and a Wigig wireless network card, not connected for security reasons The external network is only connected to 50 sets of C to form a short-range internal network. It uses 48GB RDIMM memory (4GB by 12) and 1333MHz, of which 25G of memory is virtualized into a memory disk as the R layer write cache layer. The Ramdisk read and write speeds are respectively 8GB per second and 10GB per second, 4K random read and write reaches 600MB per second, allocates 512MB of RAMDISK on the S side for each of the 50 devices on the C side, and generates an image file of RAMDISK, which is loaded when the device is powered on, and saved when the device is powered off. Although the speed of the RAMDISK is 10GB per second for S, since the actual speed of the CAT6 network is 110MB/S to 120MB/S, even after the wireless 802.11ac is connected to the grid at the same time, it is only 200MB per second, so for the C-side only The sequence reads and writes 200MB/S, 512k randomly reads and writes 200MB/S disks, and caches C writes and frequently used small files into the RAMDISK. The processor is Intel Xeon E7540 (2.0GHz, 12M cache, 6.4GT/s QPI, No Turbo), 6C, 45W, two 3D V-NAND manufactured Samsung 850 Pro 128GB solid state hard disk composed of NAND composed of Raid0, read and write The speed reaches 1GB per second, 4K random read and write reaches 160MB per second as the D layer read cache layer, the SSD solid state hard disk creates the buffer data cache file package, and the data cache package is allocated for C. The initial size of each data file is 4GB. The read operation file of C is cached in the data package. Considering that the network files and program files accessed by the C or so C terminals are highly similar, a 2GB RAMDISK cache is allocated for such a common cache. The V-cache layer divides 50 virtual private servers (VPS) dynamically allocated resources based on OpenVZ, and allocates them to different ports for C-side users to connect. The dynamic memory limit of each device is 512MB. Sharing a 24GB disk, the tool program commonly used by the company is pre-installed on the disk, and a sub-account permission control mechanism is created in the storage of the data portion of the disk. Different users have different usage and login rights, and different users have different Different files also have different permissions. The reason why the vps is divided is that the work of the employees cannot interfere with each other in the work environment. Since no display equipment is required, the total cost is about 5,000 yuan. An abstract schematic of the device has been illustrated in Figure 2.

Part L: The transmission line is mainly CAT6 Gigabit network cable, but for a real company, due to the different degrees of computer, the new computer usually has a Gigabit network card, and the old computer may still be a 100M network card, the old-fashioned for the 100M network card. The computer is connected to C via USB 3.0 to Ethernet at the terminal. Due to the limited transmission speed of CAT6, if you want to get better results, you can connect to the wireless network at the same time. For older desktop computers without a wireless network card USB to 802.11ac NIC device to obtain wireless network. The cost is around 1,000 yuan.

The acceleration structure of the first example is shown in Fig. 5.

Effect evaluation: 1. The speed brought by the cache is improved. In the original CrystalDiskMark speed test of a C-side machine hard disk, the sequence reads 75MB per second, sequentially writes 23MB per second, random 512K reads 69MB per second, and randomly writes 13MB per Second, random 4K read 4MB per second, random 4K write 5MB per second, after using this scheme, the above speeds basically reach 180MB per second or more, wherein sequential read cache and 512K random read and write basic contribution from S-side SSD data package The speed, write cache and 4K random read and write basic contribution from the S-side RAMDISK, considering the cache hit rate, the overall acceleration effect should be 2-3 times; 2. The speed and efficiency of the virtual layer performance support, enterprise environment Mainly in the office environment, there are a large number of file processing and operation of the industry software. For the work that can be executed in the background or wait for the processing result after execution, this scheme can be transferred to the S end, and other S-end features such as the S-end can also be utilized. Bandwidth and processing performance advantages can solve some corresponding work faster, or achieve the role of key work safety backup and debugging of work problems; 3. Common cache and common prefetch The network files and program files accessed by the C or so C terminals have a high degree of similarity, and processing on the same S side also significantly improves the cache hit rate.

Cost evaluation: The total cost is about 10,000 yuan, not more than 20,000 yuan, which can improve the performance of 50 C-sides.

The second example is a family-oriented, active home computing and wireless acceleration center. The home equipment will be mainly wireless in the future, so the L layout is based on wireless. In the past, households generally had PC equipment, often desktop computers. In the case of this sample solution, the PC equipment could be replaced by the S of the scheme, or by modifying the original PC equipment as the S of the scheme. The meaning of the original PC changed from Personal Computer to Personal Center.

The acceleration scheme connection diagram of the second example is shown in Fig. 6.

Part S: A Personal Center prototype with high-speed local area network, high memory and memory-based virtual disk, low storage capacity, simplified peripheral configuration, quiet design, and long-term work orientation (because general storage can be replaced by cloud storage) ), with 16G DDR3 1600 memory (4*4G), 12G of memory and virtualized into memory disk as R layer, read and write speed is 10GB per second and 12GB per second, two 64GB SANDISK SDSSDP-064G-G25 solid state The hard disk consists of Raid0, which has a read/write speed of 800MB per second and 700MB per second. It is a D layer, an Intel i3 processor, and a Dell Wireless 1601 Wigig network card ($60). The cost is about 6,000 yuan. In addition to providing a two-level cache similar to the first example, a remote VPS with screen size and resolution adaptable to device changes is also provided for the accelerated mobile device as the V layer. The remaining hardware resources that are not available for caching can be used as a traditional desktop computer.

Part L: Mobile devices such as HTC ONE phones or iPads use the 802.11ac protocol to connect to the S-end, up to 120MB per second, while the ROM reads itself up to only 17M per second, with writes of up to 8M per second. Old-fashioned notebooks and portable ultrabooks can be used to obtain high-speed wireless networks via one or more USB-to-Gigabit wireless network cards (such as the ASUS USB-AC56 adapter that converts the USB3 interface to a Gigabit wireless network card that supports the 802.11ac protocol). The new notebooks and ultrabooks themselves have Gigabit wireless network cards, and even models such as Dell's Latitude 7440 have 10 Gigabit wireless network cards.

The acceleration structure diagram of the second example is shown in Fig. 7.

The flow chart of the read operation of the second example is shown in FIG.

The flow chart of the write operation of the second example is shown in FIG.

Performance evaluation: Mobile devices, older computers, have significant performance improvements. And the V layer is functionally capable of extending the mobile device.

Cost assessment: Considering that many homes also need a high-performance desktop computer, the cost of this solution is actually an incremental cost of upgrading from a desktop computer to a wireless service S-end, and the configuration cost of L, which is about 1000. yuan.

The third example is pure cache-based acceleration, a simple active home and personal wireless acceleration prototype. The R layer and the D layer are physically merged together, and the V layer is also embedded in the D layer. An abstract schematic of the device has been illustrated in Figure 3.

The acceleration scheme connection diagram of the third example is shown in FIG.

Part S: Wireless Accelerated S-Side prototype with wireless WiGig NIC with DRAM cache (R-layer write cache layer), with four-channel SLC NAND flash to form Raid (D-layer read cache layer), and in the Raid Divide the area, create a virtualized Windows environment (including the library and registry required by the program), and virtualize the application to pre-store more program files and program system environment files in the device, more thoroughly avoiding The hard disk read and write in the program is used as the V layer, and the speed is 1 GB per second. In addition, the device can retrieve the memory of the system part together with the R layer D layer of the device to form a complex cache across devices, which compensates for the problem of write cache limitation of low wireless speed devices. As future resistive memory and phase change memory industry costs are reduced, resistive memory or phase change memory can also be used as the D-layer read buffer layer herein.

The algorithm and architecture of the device also include: 1. Intelligent compression and automatic background release for system memory; 2. Long-term monitoring and identification of user habits to determine which data is to be used by the system, pre-existing in the device; 3. Multi-channel The mode, as described above; 4. The device virtualizes the application to pre-store more or even all of the program files and system environment files required by the program in the device, as described above. (Virtualization principle is mainly to use the sandbox virtualization technology, first install the application to the running, all the actions are recorded and process the cost of the file, when the main program file is executed, it will temporarily generate a virtual environment. Execution, like the shadow system, all the operations involved are done in this virtual environment, and will not move the original system. After this processing, all the calling files are in the V layer, and will not be safe. Installed on the C side. )

Part L: Assume that the user's 2008 desktop computer has a USB AC1200 network card installed on the front two USB3s, and then the bandwidth is superimposed, 110MB per second multiplied by 2 to get 220M per second bandwidth. HTC ONE mobile phone supports 802.11ac protocol, the theory can reach 120MB per second bandwidth (actually HTC ONE is 58MB per second in current test), and its own ROM read is only 17M per second, and the write is only 8M per second.

The acceleration structure diagram of the third example is shown in FIG.

Acceleration effect: similar to the second example, but the type of acceleration is single, and the cost and power consumption are very low. Suitable for scenarios with more mobile devices. Mobile device storage devices have power consumption and volume limitations, especially mobile phone flash memory, it is impossible to do multi-channel design and complex circuit design, and the slow writing speed of the flash itself makes the mobile device itself accelerate. Performance is impossible and unnecessary. With the S device, the mobile device can achieve performance improvements when necessary.

In the case of the third example, we also added the design of the collection and feedback of the modes that different applications cache on their respective S after a period of optimization. As shown in Figure 12, S will upload these in ciphertext. Optimized cache mode configuration data in the respective systems to a processing machine, processing the statistics of various applications and related files cached by the machine on multiple C-sides, to optimize or predict the cache structure, and then feedback to Each S-end and the new S-end (for example: a certain folder of a game program on a large number of C-ends exhibits frequent read characteristics, and when S services a new C-end, if the program is found, the pre-preparation can be directly performed. The nature of the work such as caching should be frequently read and written folders on other devices to high-speed devices without re-accumulating cached data; Example 2: A large number of programs on the C-end are showing frequent write jobs, such as a shopping browser. When the browser is launched, the browser can be preliminarily assigned a larger S-side write cache layer without re-accumulating cached data.

The above three examples are all active designs, which are characterized by specially designed S and better acceleration performance. For these active-type design acceleration systems, there is the possibility of performance from the upper level of high performance to the lower level of performance with lower portability and higher portability and lower power consumption. In fact, in addition to the above examples of implementation, you can create more instances, such as the topological relationship between S and C, and possibly create more acceleration systems. In addition, in some cases, if existing equipment is used for retrofitting, and power and cost are not considered, S can be virtualized on existing large computers, servers, or high-performance computers. Virtually out the R layer D layer V layer, etc., and install the appropriate modifications to be converted into S devices, and then install the driver and L server to achieve the purpose of C acceleration.

The fourth example is a device designed specifically for system retrofits. The retrofit device has a software portion and a hardware portion. The software part is divided into a server and a server. After the server is installed on the device to be transformed into the S end, the device to be modified is re-layered to create a write cache layer and a read cache layer, and Server-side communication interaction cache Command and data, when the server is installed on the C side, it will intercept the C-side I/O for redirection, change the C-side cache structure, call the S-side cache layer as a cache, and interact with the server to exchange the cache. Form a cross-device caching system. The hardware part of the device includes two ends of the wireless transmission L, with a USB interface and a Wifi network card, and one end can be connected to the S or C by USB (including a generalized USB interface such as MicroUSB), and one end communicates with each other.

In this experimental environment, we have a 2009 assembled desktop computer with a USB3 interface and an HTC ONE phone. After the two devices are installed, the desktop caches the disk through virtual memory, and can communicate with HTC ONE with the 802.11ac wireless protocol to speed up HTC ONE. HTC ONE's storage write speed was originally only 10MB per second, but it has an 802.11ac NIC speed of up to 100 megabits per second.

BRIEF DESCRIPTION OF THE DRAWINGS:

Figure 1. Schematic of the device. At the top of the figure is a typical short-distance, multi-channel, local-performance network-based cross-device caching and computing virtualization system that provides equipment acceleration for general enterprises, homes, and individuals. The new I/O mechanism of the next computing device in the system is shown below the figure. The detailed design of the typical S devices A and B is shown in Figure 2 and Figure 3, respectively.

Figure 2. Accelerated server device designed for a wired system, as described in Case 1 of the Implementation Cases section.

Figure 3. Accelerated server device designed for wireless systems, as described in Case 3 of the Implementation Cases section.

Figure 4. The acceleration scheme connection diagram of the first example is mainly based on short-distance CAT6 wired connection and short-distance multi-channel connection to form a cross-device buffer and calculation system. At the center of the drawing is the S of the first example, and the periphery is C.

Figure 5. Accelerated block diagram of Example 1, depicting a new cross-device cache structure of the accelerated device C. C's I/O is intercepted and reallocated.

Figure 6. The acceleration scheme connection diagram of the second example is mainly based on short-range wireless connection and short-range multi-channel connection to form a cross-device buffer and calculation system. At the center of the drawing is the S of the second example, and the surrounding is C. In the counterclockwise order, the types of the C devices are smart watches, smart glasses, tablet computers, smart phones, and large-screen human-computer interaction devices.

Figure 7. Accelerated block diagram of Example 2, depicting the new cross-device cache structure of Accelerated Device C under Example 2. C's I/O is intercepted and reallocated. In the case, the accelerated device also includes many mobile device types. The mobile device is limited by power consumption and volume, and the processing performance and the cache medium itself are seriously insufficient. However, mobile devices often have higher wireless data speeds due to mobility requirements. The virtual environment can also enable the mobile device to indirectly process the Window application to obtain a corresponding operating experience.

Figure 8. Flow chart of the read operation of the sample two

Figure 9. Flow chart of the write operation of sample two

Figure 10. Acceleration scheme connection diagram for example three, this acceleration scheme is a simple scheme based on short-range wireless connection The short-range multi-channel connection constitutes a cross-device cache and computing system. At the center of the drawing is the S of the third example, and the surrounding is C. In the counterclockwise order, the types of the C devices are smart watches, smart glasses, tablets, smart phones, and desktop computers.

Figure 11. Accelerated block diagram of Example 3, depicting a new cross-device cache structure for Accelerated Device C in Example 3. Since the S device of the third example is a simplified device, there is a corresponding adjustment in the cache hierarchy system. For a detailed description, refer to the introduction of Case 3 in the implementation case section.

Figure 12. Example 3 uses a cache configuration optimization data collection, analysis and feedback mechanism based on multiple S-C systems.

Claims (20)

An external computing device acceleration method based on a server and an external cache system, comprising at least one server (hereinafter referred to as S), and the server passes the data transmission device (hereinafter, L, L may be a wired transmission type or a wireless transmission) Type or both) and the accelerated computing device (the computing device here does not specifically refer to the computer, but should include a variety of computers, tablets, mobile phones, robots or smart hardware, etc., hereinafter referred to as C), and The server works at least: S provides a relatively high-speed or high I/O performance cache device for C (such as Ramdisk or resistive memory or high-speed solid-state hard disk RAID) for C cache system and application Common files or scattered small files or network files that are frequently read and written, as a cache, transfer C to partial access to its relatively low-speed or low I/O performance devices, thereby providing C with acceleration or improving its I/O performance. A method according to claim 1, characterized in that, in addition to establishing a cache system, S also provides computational performance support for C, and the method may be any one or more of the following three modes: 1. S-shared computing task Second, through the virtual architecture and hardware resources in S, allowing C to run this virtual layer in S remote operation mode, display interface on C and realize interaction with users, one S can create virtual architecture for multiple C According to the real-time usage of C, the S-resources are deducted, and the virtual architecture separated in S can be a virtual machine or a virtual application layer; 3. The S virtualizes the application to pre-store the program files and programs required by C. The system environment is in the S device, C can run the application directly in S. A method according to claim 1, wherein S is accelerated by C for three accelerating layers, including a write buffer layer (hereinafter referred to as R Layer or R layer, such as Ramdisk, etc., S simulates memory as a disk, And in which C is created for C to get a larger cache speed), random read cache layer (hereinafter referred to as D Layer or D layer, such as resistive memory, or NAND disk array combined in RAID mode), virtual layer ( Hereinafter referred to as V Layer or V layer, by providing application virtualization or hardware virtualization to provide computing performance support for C), of course, due to power consumption and cost considerations in specific environments, the above three layers are at the physical device level. Can be enhanced, or merged (for example, for enterprise application environments, the V layer may need multiple to accommodate different application needs, while for simple home applications, the R and D layers can be physically combined into one RD layer) . A method according to claim 1, wherein between S and C, any two or more of the following topological relationships are formed: 1. S and C may be one-to-one, one-to-many, Or a many-to-many relationship, such as the family environment may be one-to-one or one-to-many relationship, and the environment of enterprises, schools, administrative units, etc. may be one-to-many or many-to-many relationships (this topology relationship is hereinafter referred to as Flexible service relationship); Second, S and C are only relative relationships, that is, C in a group relationship can be S in another group relationship, such as an accelerated computer, which can simultaneously accelerate for another computer, and correspondingly The S in a set of relationships may also be C in another set of relationships, such as the S-acceleration of the next-level performance and cache speed by the computing device of the higher-level performance and cache speed (the topology relationship is hereinafter referred to as Performance flow relationship); three, multiple C can simultaneously Mutually acting as S to form an accelerated networking of P2P to obtain performance improvement (for example, where a computer C1 has a large memory, which can provide a stronger small file caching capability and a write operation buffer, but the disk speed is relatively insufficient, and some Computer C2 has multiple large flash SSDs and increases bandwidth in RAID form. With a stronger read operation cache, C2 can be S with C1, and C2 and C1 write caches are written to C1. The read operation cache is more written to C2, which is hereinafter referred to as the performance complementarity relationship). A method according to claim 1, characterized in that the pre-stored design is further adopted: by performing long-term monitoring and recognition on the C user habit, determining which data is to be used by the C system, pre-existing in S, and C will be directly from the device. Get the data and transfer it to the memory or processor to reduce the read and write of the C hard drive. A method according to claim 1, characterized in that encryption is also provided for all cache files in S. A method according to claim 1, wherein a plurality of Ss are networked in a high-speed connection manner such as an optical fiber to enhance the capability of read-ahead analysis and to obtain higher cache performance. A method according to claim 1, wherein S combines its hard disks in a RAID manner to obtain a larger buffer speed, especially a random buffer speed. A method according to claim 1, wherein S emulates the memory as a disk and creates a cache for C therein to obtain a larger cache speed and write cache speed. A method according to claim 1, wherein when the network file and the program file accessed by the plurality of C terminals have a high degree of similarity, a shared buffer is allocated to the S-side to cache such a common buffer. A method according to claim 1, wherein the cache structure is optimized or pre-judged according to statistics of various applications and related files cached by the plurality of C-sides served by the single or multiple S pairs ( Example 1: A large number of folders of a game program on the C side exhibit frequent reading characteristics. When S services a new C-end, if the program is found, the pre-judgment work can be directly performed, such as caching in other The device is frequently read and written to the high-speed device without re-accumulating the cached data. For example, a large number of programs on the C-side display frequently write jobs, such as a shopping browser, when the browser is started. It can be preliminarily assigned a large S-side write cache layer for it. A method according to claim 1, wherein L is of a wired type: L can be an optical fiber, a Thunderbolt transmission line, an extended USB 3.0 or higher transmission line, a super five type network cable, or any other type of transmission speed greater than 60 MB per second. Wired transmission equipment, S is directly connected to C through these transmission cables; L can also be a limited type including USB Ethernet adapters (such adapters have a USB interface on one end and an Ethernet interface on the other end, through the present Some USB input ports add an Ethernet port to the computer and other necessary accessories. Use the USB port of the adapter to connect to S or C, and use a high-speed network cable between them. A method according to claim 1, wherein L is of a wireless type and L can be supported by WiGig technology. The wireless network card can also be a base, an external box or a protective cover that is connected between the wireless transmission and the server S, and then connected to the accelerated computing device C by a USB interface or a Thunderbolt interface or other interface, or any other type of inclusion. A device that wirelessly transmits parts. An apparatus for implementing the method of claim 1 wherein L is of a multi-channel type, that is, the connection between S and C passes more than one L, or L includes more than one interface to C, thereby Increase the transmission bandwidth (Example 1, C on the one hand through its own Wifi and S connection, on the one hand through the USB interface Wifi adapter is also connected with S, after the connection to merge bandwidth; Example 2, C multiple USB interfaces use USB The Wifi adapter of the interface is connected to the S, thereby increasing the Wifi bandwidth upper limit to the USB bandwidth upper limit; for example, the Wigig base is used to communicate with the S by Wigig, and at the same time, the Bluetooth and MicroUSB are connected to the C). An apparatus for implementing the method of claim 1 is characterized in that: the S cache server further adopts any one or more of the following three layers of acceleration layer specific design: First, virtualizing part of the memory into a memory disk The R layer writes the cache layer, allocates part of the R layer cache for the C-end devices, and generates an image file of each memory disk cache, which is loaded when the device is booted, and is saved when the device is powered off; 2. The solid-state hard disk or multiple solid-states at the S end A read buffer area is created in the Raid0 of the hard disk. Third, a virtual dedicated server that allocates a plurality of dynamically allocated resources to the S is allocated to different ports for use by the C-end user. An apparatus for implementing the method of claim 1, wherein S is connected to C by a wireless network, and a virtualized Windows environment (including a library and a registry required by the program) is created on the S, and The application is virtualized, so that more program files and program system environment files are pre-stored in the device, and the hard disk reading and writing in the program is more completely avoided. An apparatus for implementing the method of claim 1 is characterized in that S also calls the memory of the C-end device part together with the cache of the S device to form a complex cache across devices. An apparatus for implementing the method of claim 1, wherein S is connected to C by a wireless network, and S provides a cache layer for C in a high speed storage device including multi-channel flash memory. An apparatus for implementing the method of claim 1, wherein S is connected to C by a wireless network, and S provides a buffer layer for C with a high speed storage device with a DRAM cache. An apparatus for implementing the method of claim 1, wherein the device of the software part of the device is divided into a server and a server, and the server is installed on the device to be converted into the S end. Re-layering the device to be modified, creating a read-write cache layer, and interacting with the server to exchange cache commands and data. When the server is installed on the C-side, the I/O of the C-side is intercepted. Orientation, change the cache structure of the C-end, call the S-side cache layer as a cache (the device described in the claim is mainly designed to transform the original computing system, according to the computing system to be modified It contains the software part and the hardware part, and may also contain only the software part).

Images