TW594469B - Method to test the overall performance of server - Google Patents

Abstract

The present invention provides a method to test the overall performance of server, which comprises first selecting at least a hardware affecting the server performance as a testing object; using the related hardware of the testing object as the testing item; confirming the performance index of the testing item; proceeding pressurization test onto the testing item, and recording the variation of the performance index; and analyzing the variation situation of performance index, so as to obtain the testing conclusion. The method to test the overall performance of server provided by the present invention can set pressure onto a specific hardware individually to observe the variation index of the overall system performance, and can integrate each testing item for testing, and set different pressures onto each testing item, so as to observe the variation index of the overall system performance, and obtain the overall system performance index, which reflects the practical hardware affecting the overall performance of server system.

Description

594469

V. Description of the invention (1) [Technical field to which the invention belongs] The present invention relates to a kind of overall performance of a server and a performance measurement of a server based on a hardware level. In particular, it involves testing to test the performance of a feeder system. , And make plastic j = over-pressure to measure the bottleneck hardware. [Previous technology] of the overall performance of the θ system. Communication systems provide functions, information, and memory settings under various systems. For the server, the server [invented the server is an electric shared server and will be equipped with it. It provides centralized computing for the entire network as a network, so it is also managed by the server and the system management information server Subject content: A kind of hard disk connected to the brain for the nodes on the Internet and information routing is called network formation and personal flow, and it is effective in improving progress, rationality, etc. It is a network-capable computer. , Disks, customer site tables and data cornerstones, serving the soul of storage and office. The computer is designed to test the stability and performance of the network as a client computer. It is used to control printers, modems, and dedicated points in the network operating system. It can also provide services such as network management. . Servers play an important role in managing a large amount of data on the Internet. It seems that the use of processors, hard disks, and physical network applications has become more and more obvious. It is more important for guilt, feminine, and scalability. It's getting higher and higher. So how is the current universal concern | 义 g. Currently, many server benchmark softwares are provided on the market. They are not quasi-tests and measure the performance of different aspects of the server. One type, two and four bodies, their working principle is: run a piece of application, one has phase

V. Description of the invention (2) The corresponding test report, the number of clients, the advanced performance of the server, such as the ability to support, these indicators reflect = high speed, different speed, file access time, etc. Summer of the overall performance: the integrity of the servo is ㊣, and π can reflect the shadow ~ specific hardware. The other is designed for the performance of parts, and ^ #. Departments # and 'are required to measure the performance of a single component or subsystem. The processor, memory, and magnetic disks cannot reflect each component. It only provides some specific information. Can be good or bad. The interaction between the two, and the nature of the mutual cooperation. Therefore, in order to solve the performance test method, the present invention takes into account each of the problems, and provides a server overall effect on the system's overall performance that affects the overall performance of the system = 2 f. , The mutual cooperation performance of the router system. The hardware pressurization test reflects the servo. The present invention provides a method for performing the following steps: ° ~ Physical performance test method, including, for example, selecting the relevant hardware of the test object that affects the performance of the server as a test method. Test object; Determine the performance index of the test item, and perform a pressure test on the test item, while recording the change of the performance index; Analyze the change of the performance index to obtain a test editorial. According to the ^ method of the performance and system of the feeder system provided by the present invention, the specific hardware is pressurized at an early stage, and the overall performance index of the system is observed. 2 The influence of the specific hardware on the system is obtained. :

V. Description of the invention (3) Specific hardware, and also test the various test items together. The interactions between .1 are taken into account, and each test J Valley is measured IS # performance change indicators, so that the load is loaded into different pressures Observe the overall performance of the system related to the characteristics of the present invention. The detailed description is as follows: 仏 and the implementation ′ are shown in the drawings to make the best embodiment. Reflected in the high-speed computing power, the long-term server's overall system: food: department :, throughput f force and so on. In order to optimize server performance, the hardware is mainly important for the balance of the two systems. Cache, processor (M_ry), cache memory A / λT, hard disk (Hard Disk ), And etwo ^ k). Host processor, local memory, and 丨 / 〇 device: must work together to provide an effective working environment for the end user. If some aspect of the system becomes a bottleneck, the system The overall performance is bound to be affected. When the in-lamp servo is a performance test, its core is to determine the system performance evaluation indicators, how to pressurize the system, and how to give a performance report to reflect the specific hardware that affects the overall system performance. In view of the above problems, by designing test items for each hardware level, the number of controllable threads, the number of bytes transmitted, and pressurizing specific hardware to solve the problems encountered in system performance testing. At the same time monitoring the system Key performance parameters, such as changes in related indicators such as memory, processor, cache memory, server, file system, etc. The above parameters have a weaving system performance reports' provides hardware based user-level servers systemic

594469

Can report. A method for testing the overall performance of a server provided by the invention of the second t is total two = 2 = Figure 1 ^. [Fig. 1] is a flowchart of a method for testing $ = of the second server provided by the present invention. First, select a piece of hardware that affects the server i 2 M as a test object (step 11 0); use the relevant hardware of the test object as a test item (step lu); determine the sex index b of the test item (step 11 2); perform a pressure test on the test item, and record the change of the performance index (step 113); analyze the change of the performance index, and draw a test conclusion (step ji 4). Among them, the test items in the present invention include: CPU, memory

(Memory), hard disk (Hard Disk), network performance (Network), cache memory (Cache), network socket (Net Socket), auxiliary equipment (Anci 1 lary). Corresponding performance indicators include, CPU throughput (I / O Data bytes / sec, I / O other data bytes / sec, I / O data

operations / sec, I / O other operations / sec, etc.), CPU utilization (% processor time), processor pair length (% processor time), I / O interrupt time (i nterrupt / sec,% interrupt time) , Disk processing time (% Disk Time), Disk processing queue length (Physical Disk \ AVG.Disk Queue, Physical Disk \ AVG.Disk Queue Length), Disk processing data volume (Avg. Disk Bytes / Transfer), Disk seek times (Avg. Disk sec / Transfer), disk working time (PhysicalDisk \% Disk Time), work queue length of the server

Page 8 594469 V. Description of the invention (5) Degree (Server Work QueuesX Queue Length), memory page transfer. Frequency (Memory \ Pages / Sec), data transmission rate (server / bytes receive / sec, bytes Transmi11ed / sec , Network interface \ bytes sent / sec, bytes received / sec, Total / sec), the amount of available physical memory (Avaiable Mbytes), the percentage of cached readings that hit the cache (c0py Read Hi ts), cache clearing The rate of data to disk (Data Flush Pages / secO), and the rate at which cached content is flushed to disk (Data Flushes / sec). Among them, the data transmission rate, the total transmission rate (Total / sec) = server / bytes receive / sec + bytes

Transmi tted / sec 〇 This product selects a certain inspection system to resonate, but also can work with the same pressure and performance indicators as the graph "," Test flow and cache first "Figure 2" In the system, the test servo provided The method of the overall performance of the device can either use individual hardware as a test object, pressurize it on specific hardware, and observe the change index of physical properties to obtain the effect of the specific hardware on the system. The test items are loaded without observing the overall performance change indicators of the system, so that the overall system and the bottleneck hardware that affect the overall system performance are obtained. When setting the hardware for testing, please refer to "Figure 2" ~ "Figure 6 = 2" ~ "Figure 6" is the invention of the second = Yes, chat, hard disk, network performance, memory, δ fe, test flow chart. Test the performance of the central processing unit CPU. Please refer to the figure below for the flow chart of the physical ~ Λ / birth test. The speed of the feeder system m can fully support the performance of the system. system

Page 9 594469 V. Description of the Invention (6)

The number of main memory, speed, capacity, and number of disk drives in the system architecture are all factors that must be considered to affect the performance of the feeder system. Therefore, it is necessary to comprehensively observe the performance indicators in order to draw system performance conclusions. When the test object is the performance of the CPU, the test items are memory, hard disk, and network socket (Net socket) (step 211). After the test items are determined, the performance index can be determined (step 212). The specific index is CPU Throughput (I / O

Data bytes / sec, I / O 〇ther data bytes / sec, I / O data operations / sec, I / 〇other operati ns / sec, etc.), cpu occupancy rate (% processor time), and the length of the opposite column of the processor (System \ processor queue 1 ength, server work queues \ queue length), parameter changes of I / O interruption time (interrupt / sec, time); and then pressurize the test items, measure and write the data to all memories (The data is specified by the user), then L = f is written and read out and verified. At the same time, increase the memory buffer Ϊ = number: disk: the number of transmission bytes of the network socket, and the key, then 1 key, and 5 changes in recorded performance indicators (step 213). Main bucket size, number of threads running at the same time. The larger the data, the more ^ the index; she obtained the conclusion of the test CPU utilization index. If the CPU throughput rate is high first / look at the CPU throughput rate index and the CPU performance is good, and the blood A and ㈣ occupancy rate is low , Said low, and the cm occupancy rate is too high, it needs to be good; if the CHI throughput rate is,

Page 10 594469 V. Description of the invention (7) A value of system \ processor queue length is too high, because most of the processor time is competing for light, indicating that the CPU itself is slow; if the value of i server work queues \ queue length is too high, Indicates that the CPU is blocked due to busy network processing. If the value of the I / O interruption time is high, it indicates that there is an I / O hardware problem, causing the CPU to block. Please refer to "苐 3 图". [Figure 3] A flowchart of a hard disk performance test according to an embodiment of the present invention. When the test object is a hard disk, select the test items as memory, body, and hard disk (step 3 11). Then determine the performance index of the test item (step 312), which is the disk processing time (% Disk Time), the disk processing queue length (Physical Disk \ AVG. Disk Queue,

Physical Disk \ AVG · Disk Queue Length), disk processing data (Avg · DiskBytes / Transfer), and disk seek times (Avg. Disk sec / Transfer). During the test, certain data is written to the file, and then the file is read to compare the read and write consistency. At the same time, gradually increase the buffer size of the memory, the number of threads running simultaneously, the amount of data transferred from the hard disk, and the number of threads running simultaneously, and record the change in performance indicators (step 3 1 3). The main test key is the size of the data. The number of threads running at the same time ’The larger the data, the better, and the larger the number of threads running at the same time, the better. After the test time is over, the recorded performance indicators are analyzed to arrive at a conclusion (step 3 1 4). If the amount of data processed by the magnetic disk is high, the performance of the magnetic disk is good; if the amount of data processed by the magnetic disk is not high, check the processing time of the magnetic disk. If the disk processing time is high, check physical Disk \ AVG.Disk Queue Length ’Avg.Disk sec / Transfer, Avg.Disk Bytes / Transfer. Physical Disk \ AVG.Disk

Page 11 594469

Qu ^ ue Length counter, check the number of system requests waiting for disk access. If the value is large, you need to change the disk type. Avg Disk sec / Transfer is too high, but the performance of the magnetic disk itself is not high, which causes performance problems, and the disk needs to be replaced. Avg.Disk Bytes / Transfer value is low, because the magnetic disk itself is not efficient, causing performance problems. -

See H? And "Picture 4" in May. [Figure 4] A flowchart of a network performance test according to an embodiment of the present invention. When the test object is the network performance, first select its test item as the memory 'hard disk' network socket and cache the memory (step 41 2). Then determine the performance indicators of the test items as cpu occupancy rate (Processor \% Process〇r Time), disk working time (physicalDisk \% Disk Time), server work queue length (Server Work Queues \ Queue Ungth), memory page Exchange frequency (Memory \ Pages / Sec), data transmission rate (server / bytes receive / sec, bytes Transmitted / sec ^ network interface \ bytes sent / sec, bytes received / sec, T〇tal / sec (total transmission rate = server / bytes receive / sec

+ bytes Transmitted / sec). Then perform a pressure test. During the test, a certain amount of data is sent out through the network, and then the feedback data is received. At the same time, gradually increase the memory, the buffer memory of the cache memory, the number of running threads, the hard disk, and the transmission data of the network socket: ί The number of running threads, while recording changes in performance indicators (Ν 413). The main test key is the data size and the ^ process ^ running at the same time. The bigger the data is, the better the number of threads running at the same time is. The analysis of the obtained performance index records is performed (step 414)

Page 12 594469 V. Description of Invention (9). By examining the network performance, you can also view the CPU usage, disk working time \ memory page exchange frequency, data transfer rate, and server's work queue length. When the total transfer rate decreases, if the page swap frequency increases significantly, it is because the memory is too small to affect the efficiency. When the CPU usage value is high, it is the efficiency affected by the CPU performance. When process〇r \% Interrupt

Time, Processor \ Interrupts / sec values are too high due to poor network card hardware performance. In this case, you should also examine server / bytes receive / sec, bytes transmitted / sec, network interface \ bytes sent / sec, and bytes received / sec values. The low value of server / bytes receive / sec indicates that the processing capacity of the local network card is insufficient, and the low value of network interface indicates that the network performance is insufficient. Please refer to the diagram "Fig. 5", which is a flowchart of a memory performance test according to an embodiment of the present invention. The performance of the memory directly affects the performance of other components. The performance of the memory is mainly affected by several aspects. The first is the reading speed, the second is the increase in memory capacity and the reading and writing of adjacent memory. Speed is matched with high-speed memory components. When the test object is memory, the selected test items are memory, hard disk, and cache memory (step 5 11). Determine the corresponding test index (step 5 1 2) as the page switching frequency (Memory \ Pages / Sec), the amount of available physical memory (Available Mbytes). Test the processor's ability to grab (snoop) and refresh (f 1 ush) instruction prefetch queues during data writing. The compression method used for the pressure test of the test items is to increase the buffer size of the memory and cache memory, the amount of data transferred on the hard disk, and the running thread.

Page 13 5. The number 1 of the description of the invention (10), and § the change of the recorded performance index +, the key is the buffer memory of the cache memory /). The larger the main test punch area is, the better the number of threads running at the same time is. The slower the change of the test index is, the better the performance is. According to the record, the page exchange frequency and the delicious body = two. The better the performance of the display memory, the more the frequency of the page change, the better the performance of the display memory. For the available physical memory, please refer to "Figure 6", "筮 R 闰, ^ Memory performance test Flow chart. Yi? 2: Mao Ming's example changes the cache logic. The test object is the cache memory. Physical :: Key: Where is it? Niu Suchuan, two J's hidden body, auxiliary equipment ( Ancillary) Determine the test index for the two U test items (Step 6 1 2) Slow: == Cache Copy Read Percent (C0Py Read Hits) Store / Month layer to page rate on magnetic disk (]) ata = ge B / saO), the rate at which the cached content is cleared to the disk (dsec — sec) is mainly performed by periodically clearing the contents of the cache memory. The verification speed of the memory data and the content of the cache memory is tested. The size of the compression method of the cache memory when performing the compression test And the number of running threads, and gradually decreases prepared Han = refresh time, while recording changes in the performance index (step 613 reads). The main test key is to set the refresh time and the number of threads running at the same time. The shorter the refresh time, the better the number of threads running simultaneously. Then analyze the performance index, when examining the cache memory, mainly check the cache copy read percentage that hits the cache, the cache is cleared to the disk, and the number of pages will quickly report to the second class indicators. The above three indicators are cached in the second cache Copy read people% good. When the rate index of the cache clearance to the magnetosphere to the magnetic disk is low, the test method of item pressure, the impact of the hardware on the system. Similarly, starting from “reducing” to all test items, the overall bottleneck hardware of the system is obtained. Among them, the preferred embodiments are only: That is, the scope of all patents for creation according to this creation is covered by the scope of creation patents. V. Description of the invention (11) The clearance and cache contents on the magnetic disk reflect the hit rate and transmission rate performance. 'When the percentage is south, the page rate on this chip, the cache, and dry cache contents are cleared', which means that the cache memory performance is good. … Dead; it is the change of the test of specific hardware, to obtain a specific pressure, to record the energy indicators of all test items, to change the energy indicators to μ ^, and to limit the performance of the shirt system. The scope of implementation of all works made by J. Temple meals: transformation and modification, both

594469 Brief description of the diagram. Figure 1 is a flowchart of the overall performance of the server provided by the present invention. Figure 2 is a flow chart of the CPU performance test of the embodiment of the present invention. Figure 4 shows the flow of the network performance test of the embodiment of the present invention. Figure 5 is the performance test of the memory of the embodiment of the present invention, and Figure 6 is the performance test of the cache memory of the embodiment of the present invention. ^ Figure 0 丨 Test method diagram; [Picture, I diagram, ί Process diagram; ξ flow [schema step 1 1 0 step 1 1 1 step 1 1 2 step 1 1 3 step 1 1 4 step 2 1 1 step 212 step 213 step 214

No.] Select at least the object that affects the performance of the server. Use the relevant hardware of the test object to determine the performance index of the test item. Pressurize the test item, analyze the change of the performance index with the change, and the test item is memory. The performance indicators of the physical, hard disk, and network are the CPU throughput, CPU column length, and I / O interruption time. The number of threads in the memory buffer, hard disk, and network socket is gradually increased and recorded. Performance analysis CPU throughput rate, CPU usage rate When a piece of hardware is a test item, the test path socket occupancy rate, small, changes in the amount of data lost during operation, and the processor as a test indicator are concluded. The number of threads and the queue length at the same time 594469 The diagram briefly explains step 311 step 312 step 313 step 31 4 step 411 step 412 step 4 1 3 step 41 4 step 511: JI /. The change of indicators such as the interruption time, the test items are memory and hard disk, and the index is magnetic disk processing time, magnetic disk processing, magnetic disk processing data volume, and magnetic disk seek time 2 gradually increase the memory. The size of the buffer, "^ 66 hard disk, the amount of data transmitted by the network socket, and the simultaneous traffic: and record the changes in performance indicators my reasoning ... length, magnetic disk to get the measurement: on the brother time and other indicators The changes are; test items are memory and hard drive memory test items ^ #connector, cache memory performance index is the length of the work queue occupied by the CPU, the processor, the memory: time? The page parent changes the frequency, the data gradually increases the memory, the number of fast-running threads, the buffer size of the memory, the amount of data, and the simultaneous running records. And record the analysis of performance indicators such as CPU utilization, magnetic work queue length, production shift time, server's work rate, and other indicators. Change the page frequency, the data transmission speed test item is memory /, 2 test conclusions' Way socket, hard drive 594469 Step 5 1 2 The performance index is the page swap frequency and the amount of available physical memory. Step 5 1 3 Gradually increase the memory, the buffer size of the cache memory, the number of threads running at the same time, and the hard disk transfer. The amount of data and the number of threads running at the same time, and record changes in performance indicators. Step 5 1 4 Analyze performance indicators such as page exchange frequency and the amount of available physical memory to draw test conclusions. Step 6 11 Test items are memory and cache memory. Body, hard disk, auxiliary equipment

Step 6 1 2 The performance indicators are the percentage of cache copy reads that hit the cache, the rate at which the cache is cleared to the disk, and the rate at which the cache is cleared to the disk. Step 6 1 3 Gradually increase the buffer size, the number of threads running simultaneously, And reduce the refresh time of auxiliary equipment test items, while recording changes in performance indicators. Step 6 1 4 Analyze performance indicators such as the percentage of cache copy reads that hit the cache, the number of pages that the cache clears to disk, and the rate that the cache clears to disk. 'Draw test conclusions

Page 18

Claims (1)

594469 VI. Application for Patent Scope 1. A method for testing the overall performance of a server, including the following steps: selecting at least one piece of hardware that affects the performance of the server as a test object; using the relevant hardware of the test object as a test item; determining the Test the performance indicators of the items, perform pressure test on the test items, and record the changes in the performance indicators; and analyze the changes in the performance indicators to draw the test conclusions. 2. The method for testing the overall performance of a server as described in the scope of application patent 1, wherein the test object may be selected from a CPU, a memory, a hard disk, a network performance, and a cache memory. Any one of a combination of caches. Q. The test method for the overall performance of the server as described in the scope of patent application 1 'where the test item can be selected from the group consisting of CPU, memory, hard disk, network performance, and cache memory Any one of a combination of a cache (Cache), a network socket (Net Socket), and an auxiliary device (Anc i 11 ary). 1. The method for testing the overall performance of a server as described in the scope of application patent 1, wherein if the test object is a CPU, the test item is a memory ^ em〇ry), a hard disk (Hard Disk), and a network socket (Net Socket). 5. The test method for the overall performance of the server as described in the patent application scope 4, wherein the test item is pressurized by adding the buffer size (Mem〇ry) at the same time ^ disc Page 19 594469 VI. Scope of Patent Application --- (Hard Disk), the amount of transmission data of the network socket (NetSocket), and the number of running threads. 6. The test method for the overall performance of the server as described in the patent application scope 4, wherein the performance indicators of the test items include cpu throughput rate, cpu occupancy rate, processor queue length, and 1/0 interrupt time. 7. As described in the method for testing the overall performance of the server as described in the scope of patent application 1, where the right test object is Hard Disk, the test items are Memory and Hard Disk. Disk). 8. The test method for the overall performance of the server as described in the scope of the patent application 7 ', wherein the test item is pressurized by increasing the size of the buffer size of the memory (Memore) at the same time as the hard disk. (fard Disk) The amount of data transferred, and the number of threads running simultaneously. 9. The test method for the overall performance of the server as described in the scope of patent application 7, wherein the performance indicators of the test items include magnetic disk processing time, magnetic disk processing queue length, magnetic disk processing data amount, and magnetic disk seek time. 10. The test method for the overall performance of the server as described in the scope of the patent application, if the test object is network performance (N etw rk), the test items are memory, hard disk ), Network socket (NetSocket), cache memory (Cache). 11. The test method for the overall performance of the server as described in the scope of the patent application 10, wherein the test item is pressurized by simultaneously increasing the buffer size of the memory and cache. Size, Hard Disk, Network Socket 594469 6. Scope of patent application-(Netscket) The amount of data transmitted and the number of running threads. For example, the test method for the overall performance of the server described in the patent range 10, the performance of the test item refers to the working time of the disk, the CPU usage, the length of the server's work queue, the memory page exchange frequency, Data transfer rate. The test method for the overall performance of the server as described in the patent application scope of claim 1, wherein if the test object is a memory, the test items are a memory, a hard disk, and a cache memory. Body (Cache). The method for testing the overall performance of the server as described in the scope of patent application 13, wherein the test item is pressurized by increasing the buffer (Buf fer) of the memory and cache. Size), the amount of data transferred by the hard disk (Hard Disk), and the number of running threads. The method for testing the overall performance of the server as described in the scope of patent application 13, wherein the performance index of the test item includes the memory page exchange frequency and the amount of available physical memory. The method for testing the overall performance of the server as described in the scope of patent application 1 'wherein if the test object is a cache, the test item is a memory (M em ry), a hard disk (Hard) D isk), cache memory (Cache), auxiliary equipment (Anci 11 ary). The method for testing the overall performance of the server as described in the scope of application patent 16 'wherein the test item is pressurized by gradually increasing the size of the buffer (Buffer Size) and the number of running threads, and each 12 13 14 15 16 17 Page 21 594469 6. Scope of patent application Gradually reduce the refresh time of auxiliary equipment (Ancillary). 18. The method for testing the overall performance of a server as described in the scope of the patent application 16, wherein the performance indicators of the test item include: hitting the cached copy reading of the cache to take the white percentage ratio, the rate of pages cleared from the cache to the disk, And the rate at which cached content is cleared to disk. Page 22