Hp COMPAQ PROLIANT 6500, COMPAQ PROLIANT 4000 Disk Subsystem Performance and Scalability

[September 1997]

.....................

............

.................

...

..............

....................

Prepared By Microsoft Windows NT Integration Team

Compaq Computer Corporation

C

ONTENTS

Disk Subsystem Overview

Disk-Related Performance Characteristics

Like Drive Scalability

Like Capacity Scalability

Disk Controller Scalability

Performance Measurement Tools

Preventing Data Loss while Maintaining Performance

Disk Subsystem Summary of Findings

ECG025.0997

17

19

24

27

28

29

W

.

Disk Subsystem Performance and

.

Scalability

.

In today's networking environments, the disk subsystem is a key element in determining

.

overall system performance. The goal of this paper is to provide informative test results

.

and performance-related information for various disk subsystems, to assist systems

.

engineers and network administrators in making decisions on disk subsystem installation,

.

optimization, and configuration.

.

3

4

This white paper also provides information on using Fault Tolerance to prevent data loss,

.

while maintaining system performance. Finally, this paper provides a section discussing

.

the advantages and disadvantages of RAID technology.

.

Help us improve our technical communication. Let us know what you think about the

.

technical information in this document. Your feedback is valuable and will help us structure

.

future communications. Please send your comments to: CompaqNT@compaq.com

.

1

HITE

APER

P

ECG025.0997

HITE PAPER

W

.

N

OTICE

.

The information in this publication is subject to change without notice.

.

C

OMPAQ COMPUTER CORPORATION SHALL NOT BE LIABLE FOR TECHNICAL

.

OR EDITORIAL ERRORS OR OMISSIONS CONTAINED HEREIN

.

INCIDENTAL OR CONSEQUENTIAL DAMAGES RESULTING FROM THE

.

FURNISHING

.

This publication does not constitute an endorsement of the product or products that were

.

tested. The configuration or configurations tested or described may or may not be the

.

only available solution. This test is not a determination of product quality or correctness,

.

nor does it ensure compliance with any federal, state or local requirements. Compaq

.

does not warrant products other than its own strictly as stated in Compaq product

.

warranties.

.

Product names mentioned herein may be trademarks and/or registered trademarks of

.

their respective companies.

.

Compaq, Contura, Deskpro, Fastart, Compaq Insight Manager, LTE, PageMarq,

.

Systempro, Systempro/LT, ProLiant, TwinTray, ROMPaq, LicensePaq, QVision, SLT,

.

ProLinea, SmartStart, NetFlex, DirectPlus, QuickFind, RemotePaq, BackPaq, TechPaq,

.

SpeedPaq, QuickBack, PaqFax, Presario, SilentCool, CompaqCare (design), Aero,

.

SmartStation, MiniStation, and PaqRap, registered United States Patent and Trademark

.

Office.

.

Netelligent, Armada, Cruiser, Concerto, QuickChoice, ProSignia, Systempro/XL, Net1,

.

LTE Elite, Vocalyst, PageMate, SoftPaq, FirstPaq, SolutionPaq, EasyPoint, EZ Help,

.

MaxLight, MultiLock, QuickBlank, QuickLock, UltraView, Innovate logo, Wonder Tools

.

logo in black/white and color, and Compaq PC Card Solution logo are trademarks and/or

.

service marks of Compaq Computer Corporation.

.

Other product names mentioned herein may be trademarks and/or registered trademarks

.

of their respective companies.

.

U.S.A.

.

Microsoft, Windows, Windows NT, Windows NT Server and Workstation, Microsoft SQL

.

Server for Windows NT are trademarks and/or registered trademarks of Microsoft

.

Corporation.

.

Disk Subsystem Performance and Scalability

.

First Edition (September 1997)

.

Document Number: ECG025.0997

.

2

(cont.)

,

PERFORMANCE, OR USE OF THIS MATERIAL

,

NOR FOR

.

ECG025.0997

HITE PAPER

W

.

ISK SUBSYSTEM OVERVIEW

D

.

Key components of the disk subsystem can play a major part of overall system

.

performance. Identifying potential bottlenecks within your disk subsystem is crucial. In

.

this white paper we identify and discuss in detail disk-related performance characteristics

.

that can help you understand how latency, average seek time, transfer rates and file

.

system or disk controller caching can affect your disk subsystem performance. Once we

.

discuss all of the disk measurement terms, we use those definitions to address

.

performance issues in each scalability section of this document. The different scalability

.

sections discussed are as follows:

.

• Like Drive (similar hard drive scalability)

.

• Like Capacity (similar drive capacity scalability)

.

• Disk Controller (multiple controller scalability)

.

This document provides disk subsystem recommendations, based on testing in the

.

Integration Test Lab of hardware and software products from Compaq and other vendors.

.

The test environment that Compaq selected might not be the same as your environment.

.

Because each environment has different and unique characteristics, our results might be

.

different than the results you obtain in your test environment.

.

Test Environment

.

The following table describes the test environment used for the disk subsystem

.

performance testing. This table displays both the one and two controller test

.

configurations that were used in the Compaq ProLiant 5000 during testing.

.

Environment Equipment Used

.

Server Hardware Platform Compaq ProLiant 5000

.

3

Memory 128 MB Processors (4) P6/200 MHz 512k secondary cache Network Interface Controllers (2) Dual 10/100TX PCI UTP Controller (4 network segments) Disk Controllers (1 or 2) SMART-2/P Controllers Disk Drives 2, 4, or 9 GB Fast-Wide SCSI-2 drives Number of Drives up to fourteen 2.1, 4.3, or 9.1 GB Fast-Wide SCSI-2 drives Boot Device (1) Fast-Wide SCSI-2 drive off the Embedded C875 controller

(cont.)

Table 1:

Disk Subsystem Testing Environment

ECG025.0997

HITE PAPER

W

.

Environment Equipment Used

.

Server Software Configuration Microsoft Windows NT Server version 4.0

.

D

.

Before beginning our discussion on disk subsystem performance, Table 2 lists the general

.

terms used in the industry to describe the performance characteristics of disk

.

performance. These general terms describe characteristics that can impact system

.

performance, so it is important to understand the meaning of each term and how it could

.

affect your system.

.

4

Service Pack 2 Compaq Support Software Diskette 1.20A Client Configuration Compaq Deskpro 575

NetBench 5.0 Test Configuration Disk Mix

Work Space 15 MB Ramp Up Time 10 seconds Ramp Down Time 10 seconds Test Duration 120 seconds Delay Time 0 seconds Think Time 0 seconds

ISK-RELATED PERFORMANCE CHARACTERISTICS

Terms Description

Seek Time The time it takes for the disk head to move across the disk to find a

Average Seek Time The average length of time required for the disk head to move to the

(cont.)

Table 1:

(cont.)

Disk Subsystem Testing Environment

Netelligent 10/100 TX PCI UTP Controller and MS-DOS

Table 2:

Disk Performance Measurement Terms

particular track on a disk.

track that holds the data you want. This average length of time will generally be the time it takes to seek half way across the disk.

ECG025.0997

HITE PAPER

W

.

Terms Description

.

Latency The time required for the disk to spin one complete revolution.

.

Average Latency The time required for the disk to spin half a revolution.

.

Average Access Time The average length of time it takes the disk to seek to the required

.

Transfer Rate The speed at which the bits are being transferred through an

.

Concurrency The number of I/O requests that can be processed simultaneously.

.

RPM (Revolutions Per Minute) The measurement of the rotational speed of a disk drive on a per

.

Table 2 lists the definitions of disk-related performance characteristics. Let’s now use

.

those definitions in the next several sections to address how adding drives to your system

.

can affect performance.

.

Seek Time and Average Seek Time

.

Seek time describes the time it takes for the disk head to move across the disk to find

.

data on another track. The track of data you want could be adjacent to your current track

.

or it could be the last track on the disk. Average seek time, however, is the average

.

amount of time it would take the disk head to move to the track that holds the data.

.

Generally, this average length of time will be the same amount of time it takes to seek half

.

way across the disk and is usually given in milliseconds.

.

5

(cont.)

Table 2:

(cont.)

Disk Performance Measurement Terms

track plus the amount of time it takes for the disk to spin the data under the head. Average Access Time equals Average Seek Time plus Latency.

interface from the disk to the computer.

minute basis.

ECG025.0997

HITE PAPER

W

.

One method to decrease seek time is to distribute data across multiple drives. For

.

instance, the initial configuration in Figure 1 shows a single disk containing data. The

.

new configuration reflects the data being striped across multiple disks. This method

.

reduces seek time because the data is spread evenly across two drives instead of one,

.

thus the disk head has less distance to travel. Furthermore, this method increases data

.

capacity because the two disks provide twice the space to store data.

.

6

Initial Configuration

New Configuration

Seek Time

Figure 1: Average Seek Time and Capacity

You can use this same concept and apply it to many different configurations. For example, if you currently have a two-disk configuration but you want to decrease the average seek time yet increase the disk capacity, you can configure a striped set of disks using four disks instead of two. This concept applies to any configuration (odd or even number of disks) as long as you are adding more drives to your stripe set.

(cont.)

Data

Seek Time

Data

Seek Time

Unused space Unused space

Unused space

Data

ECG025.0997

HITE PAPER

W

.

Average Latency

.

Manufacturers have built and continue to build hard disks that spin at designated rates. In

.

the early years of the personal computer (PC) industry, hard disks on the market could

.

spin at approximately 3600 RPMs. As the market demand for better system performance

.

increased, disk manufacturers responded by supplying faster spin rates for hard disks.

.

By producing faster spinning disks, manufacturers reduced the amount of overall access

.

time. Average latency directly correlates to the spin rate of the disk drive because it is, as

.

defined earlier in Table 2, the time required for the disk to spin half a revolution.

.

Therefore, this direct relationship in improving hard disk spin rates can contribute to better

.

system performance by reducing the average latency on a disk.

.

Manufacturers understand the need for better system performance and continue to

.

provide new and improved hard disks. With today’s hard disks spinning at 7200

.

revolutions per minute (RPMs) and the hard disks of tomorrow spinning at the rate of

.

10,000 RPMs, we can see that manufactures continue to address the issue of faster

.

performance. Table 3 provides a brief history on hard disks listing spin rates, disk

.

capacities available and approximate dates the disks were available to the market.

.

Disk Spin Rate Disk Capacity Approximate Date Used

.

3600 RPMs Up to 500 MB 1983 – 1991

.

4500 RPMs 500 MB – 4.3 GB 1991 – Present

.

5400 RPMs 500 MB – 6 GB 1992 – Present

.

7200 RPMs 1 GB – 9.1 GB 1993 - Present

.

10,000 RPMs 4.3 GB and 9.1 GB 1997 - Present

.

Now that we discusse d the direct relationship between disk spin rates and system

.

performance, let’s examine how drive scaling can affect latency. In Figure 2 - Example 1,

.

the initial configuration shows the disk has to spin halfway around before the disk head

.

can start to retrieve data from sector 5. In the new configuration, the disk has to spin half

.

the distance than before to retrieve the same data. Thus, the latency time has been cut

.

in half.

.

7

(cont.)

Table 3:

Hard Disk History

: Remember that the disks

Note

used in Figure 2 ar e ide n tic al in size and RAID configuration.

ECG025.0997

HITE PAPER

W

.

However, be aware that the average latency time might not always decrease when addi ng

.

more drives to your system. For example, in Figure 2 - Example 2, the new configuration

.

shows that the amount of time it takes to retrieve the data from sector B is actually longer

.

than the initial configuration. The reason for this is that the disk has to spin half way

.

around to read sector B. In the initial configuration the disk only had to spin one-eighth a

.

revolution to read the identical data. But, keep in mind that the initial configuration for

.

Example 2 required both seek time and latency time.

.

8

Initial Configuration

New Configuration

Example 1

Figure 2: Average Latency

Overall, these examples show us that in some configurations, as shown in our first example, drive scaling would be a definite performance advantage. However, in other configurations it is not clear if you receive a performance gain because of the components involved, such as the combination of average seek time and average latency time used in Figure 2 – Example 2.

When you combine these terms (average seek time and average latency time), you define another disk measurement called average access time, which is discussed in the upcoming section. From the information provided in this section, we know seek time plus latency (or average access time) is a key in determining if performance is truly enhanced in your system.

(cont.)

5

3

Example 1

A

7

1

G

Disk Head

4

D5E

3

C

B

2

H

A

1

Disk Head

C

Example 2

E

6

F

G

7

Example 2

8

B

8

6

4

2

Disk Head

D

F

H

ECG025.0997

HITE PAPER

W

.

Average Access Time

.

Average access time is simply described as average seek time plus latency. What this

.

equates to is the amount of time the disk has to seek to find the data plus the time it takes

.

for the disk to spin under the head. For example, Figure 3 contains a disk with two tracks

.

of data on it. Track 1 contains data sectors 1 – 8. Track 2 contains data sectors A – H.

.

Thus, in our example, the disk head has to move (or seek) from the current position (track

.

1, sector 1) to the track you want to read (track 2, sector C).

.

For the purpose of our illustration, Figure 3 displays the disk head performing these

.

functions separately. However, in reality the disk drive performs both seek and latency

.

functions simultaneously.

.

Figure 3: Average Access Time

.

Transfer Rates

.

A disk subsystem is made up of multiple hardware components that communicate by

.

transferring data to and from the disk(s) to a computer. The main parts of a disk

.

subsystem are as follows:

.

• Hard Disks

.

• SCSI Channel

.

• Disk Controller

.

• I/O Bus

.

• File System and Disk Controller Caching

.

9

(cont.)

Average Seek Time + Latency = Average Access Time

Data

5

E

6

F

G

7

H

8

Unused space

Latency

3

2

Seek Time

4

D

C

B

A

1

Disk Head