Intel Server RAID Q1 2009 Bootcamp Training Student Workbook Enterprise Products and Services Division
Intel® Server RAID Boot Camp Training Q1 2009 Disclaimers The information contained in this document is provided for informational purposes only and represents the current view of Intel Corporation (“Intel”) and its contributors (“Contributors”), as of the date of publication. Intel and the Contributors make no commitment to update the information contained in this document, and Intel reserves the right to make changes at any time, without notice. THIS DOCUMENT IS PROVIDED “AS IS.
Q1 2009 Intel® Server RAID Boot Camp Training RAID Products and Technologies Workshop Overview Objective This workshop is designed to provide the student with - An understanding of the correct methods of measuring RAID performance - An understanding of the performance impact of RAID cache settings This lab workbook contains the following labs: Lab 1: Correct Methods of Measuring RAID Performance Lab 2: Impact of Cache Settings on RAID Performance Course Pre-requisites None Course Length The estim
Intel® Server RAID Boot Camp Training Q1 2009 Lab 1: Correct Ways of Measuring RAID Performance Introduction It is a common mistake to measure RAID performance using wrong tools or with wrong test settings. Results of such tests can be very misleading. When measuring the RAID performance, it is important to understand how the measurement tool works, the capabilities of the tool, and any limitations.
Q1 2009 Intel® Server RAID Boot Camp Training Performing the Lab This lab consists of the following parts: A. Preparing for the lab B. Comparing sequential and random performance C. Comparing performance with different I/O queue depths D. Comparing performance with different transfer block sizes * If you are using our ‘RAID Products and Tech Workshop Overview V1.0.exe’ tool to simulate lab 1, click ‘>>Lab 1: Correct Ways of Measuring RAID Performance’ button as shown in Figure 1. Figure 1.
Intel® Server RAID Boot Camp Training Q1 2009 A: Preparing for the lab 1. Clear RAID configuration a. Run ClearVD.bat file located on the desktop to clear partition information from the RAID array that may have been left behind from other labs. b. Open the Intel® RAID Web Console 2 by clicking the shortcut on the desktop. c. Select Intel® RAID Controller SRCSATAWB. d. On the menu, select Operations -> Configuration –> Clear Configuration 2. Create RAID 5 array a.
Q1 2009 Intel® Server RAID Boot Camp Training B: Comparing sequential and random performance Sequential performance and random performance are two different characteristics of a storage subsystem. The difference comes from the mechanical nature of HDDs. Both characteristics are important. Some applications may require higher sequential performance, other applications may require higher random performance, and some applications may require both.
Intel® Server RAID Boot Camp Training Q1 2009 C: Impact of I/O queue depth on performance I/O queuing is important for getting maximum performance out of the RAID subsystem. It allows sending I/Os to several HDDs in parallel. It also allows getting maximum performance out of individual HDDs by optimizing heads movement. Queue depth (or # of Outstanding I/Os) is the number of I/Os that can be issued by an application in parallel before the first I/O is completed.
Q1 2009 Intel® Server RAID Boot Camp Training D: Impact of transfer block size on performance Every data transfer request has a certain size. This size may be equal to the file size in case of small files, or it may be equal to a database record size, or it may be hard-coded in an application.
Intel® Server RAID Boot Camp Training Q1 2009 Summary A. Sequential performance is measured in MB/s; Random performance in measured in IOPS. B. I/O queuing is important for getting maximum performance – use IOmeter. C. Transfer block size has big impact on both sequential and random performance.
Q1 2009 Intel® Server RAID Boot Camp Training Lab 2: Impact of Cache Settings on RAID Performance Introduction Incorrect cache settings are the most frequent reason for slow RAID performance. Depending on the measurement test, performance can be up to 50 times slower with incorrect cache settings than with correct cache settings. In this lab, we demonstrate the impact of the following settings: Read Policy, Write Policy, and Disk Cache Policy.
Intel® Server RAID Boot Camp Training Q1 2009 Performing the Lab This lab consists of the following parts: A. Measuring impact of Read Ahead mode on sequential read performance B. Measuring impact of write caching on sequential write performance * If you are using our ‘RAID Products and Tech Workshop Overview V1.0.exe’ tool to simulate lab 2, click ‘>>Performance Lab 2: Impact of Cache Settings on RAID Performance’ button as shown in Figure 3. Figure 3.
Q1 2009 Intel® Server RAID Boot Camp Training A: Measuring impact of Read Policy on sequential read performance Intel often gets complaints about slow RAID performance when copying large files. Copying large files, under both Windows and Linux, is done as a single threaded process with no I/O queuing and with transfer block size of 64K. As we have seen in Labs 1C and 1D, this type of access results in much slower than the maximum read speed.
Intel® Server RAID Boot Camp Training Q1 2009 B: Measuring impact of write caching on sequential write performance Write cache settings have a very big impact on write performance. With RAID 5/6/50/60, it is important to set the Write Policy to Write Back mode with any type of access – sequential or random, with or without I/O queuing. Otherwise, performance will be much slower. With RAID 0/1/10, the optimal settings depend on the application and its access pattern.
Q1 2009 Intel® Server RAID Boot Camp Training e. Set Disk Cache Policy to Disabled. f. Click Go. g. In IOmeter, click the Results Display tab, then click the Green Flag button to start the test and wait for 10 seconds for the test to complete. h. Record the Total MBs per Second result into the corresponding cell in the table below. 4. Measure sequential write performance with Write Back and Disk Cache Enabled a. In the RAID Web Console, click the Logical tab and select Virtual Disk 0. b.
Intel® Server RAID Boot Camp Training Q1 2009 Answers Lab 1-1. If we change 8 KB random read pattern to 16 KB random read, which performance metric changes more – IOPS or MB/s? There is almost no change in IOPS, because reading 16 KB versus 8 KB adds very small delay compared to moving heads from one location to another. Performance measured in MB/s is almost double, because with almost the same amount of IOPS, the amount of data transferred in every I/O is 2X times larger. Lab 1-2.
Q1 2009 Intel® Server RAID Boot Camp Training Reference Performance Results Lab 1B IOPS MB/s 64 KB Sequential Read ___6627.44______ IOPS __414.22_______ MB/s 64 KB Random Read __546.70_______ IOPS ___34.17______ MB/s Lab 1C Outstanding IOs MB/s 1 __127.17________ MB/s 2 ___231.49______ MB/s 4 __310.30_______ MB/s Lab 1D Transfer Block Size MB/s 8 KB __42.89________ MB/s 64 KB ___127.72______ MB/s 1 MB __379.89_______ MB/s Lab 2A MB/s Adaptive Read Ahead __306.