1. Test Machine
Test Methodlogy - Page 1
- Introduction
This article has the propose to explain how our tests are performed
and give some additional information that doesn't covered at our hardware reviews.
Our main interest when designing the test platform was to have the ultimate
performance among with the higher stability. The tests of hard disks is complicated
issue and our tests try to unveile each drive's performance and any possible
problems.
- Test Machine
For our tests we chosen Intel's P4 platform. Our test machine
assempled from the following parts
- MotherBoard: AOpen
AX4B Pro
- CPU: P4 at 1.5Ghz
- Memory: 512MB DDR266
- CPU Cooler: AOpen
AC715E
- SVGA: AOpen
GF3-DV
- Case: AOpen
H600A (400W)
- Mouse: AOpen
W-23G
- Network: AOpen
AON-325
- Sound: AOpen
AW320
- SCSI Controller: Adaptec
29160 (firmware 3.10)
- Promise
ATA133 Controller
- CD-RW Drive: AOpen
CRW3248 (firmware 1.00)
- DVD+RW: AOpen
RW5120A (firmware 1.37)
- FireWire Card (IEE1394)
All tests are performed under WinXP Pro operating system with
all latest fixes/patches. All disks are installed as MASTER in the secondary
IDE Bus. At the Primary IDE Bus IBM's DTLA 30730 (ATA100) disk is being used.
All tested drives are logical fixed drives ,under DOS 6.22, with Fdisk command.
No partitions are specified. All cables are ATA66/100/133 certified in order
to have the maximum performance.
2. SCSI Mechanic
Test Methodlogy - Page 2
Measurement software
- SCSI Mechanic
SCSI Mechanic 3.0x is used to perform LINEAR reading/writing tests.
Under the Exercise Media tab we find the following picture:
In this mode both Reading and writing tests can be performed. In first we are
using the "Read Only" test mode option and we perform in row all 3
tests:
a) Random
b) Sequential
c) Butterfly
The Kbytes/transfer are settled in default value (64kb/s). The time which each
test last depends from the disk drive's total sectors since the test starts
and counts the total sector operations. When SCSI Mechanic reaches the outer
limit of the drive, it stop the reading/writing test and we take the average
Kb/s test number:
As we said earlier also writing tests are performed. We select trough the pull
down menu "Write Only" and we make all tests in ROW, 3 times repeadetly:
After you press ok the software prompts that this test will destroy all data
(and partitions) in the select drive:
As long the test continues, it gives back information about the drive's writing
performance in the particular test and we track down the average Kbytes/sec
value, when the test completed:
3. WinBench
Test Methodlogy - Page 3
- WinBench 99
Ziff Davis WinBench 99 v2.00 is the most famous HD measuring software. The
software can show us the transfer rate throughput the whole HD and other interesting
results and of rank each disk with a "WinMark" value. For the WinBench
tests, each HD is formatted in one single partition under NTFS file system.
All tests are performed 3 times and posted numbers are the average of for each
HD:
4. IPEAK
Test Methodlogy - Page 4
- Intel's IPEAK Performance Tools v3.0
The Storage Performance Tool (SPT) consists of five utilities that measure
and analyze disk usage in the system:
-
AnalyzeDisk: Checks the disk operation independently of the OS or drivers,
and reports on the disk surface, sectors, etc.
- Win32 Tracing Kit: Runs a trace of disk activity (reads and/or writes)
and stores the data in a binary file for later analysis by AnalyzeTrace, AnalyzeLocality,
or RankDisk. The Win32 Tracing Kit does not display the output data.
- AnalyzeTrace: After running Win32 Tracing Kit, use this utility to
analyze and display most of the data available from the disk trace.
- AnalyzeLocality: After running Win32 Tracing Kit, use this utility
to analyze and display the locality information from the disk trace. Locality
is an analysis of the distance (in space and/or time) between disk accesses.
- RankDisk: Can run standard disk benchmarks (workloads)
built into the tool or can mimic an existing trace made on one disk onto another
disk. In either case, the tool graphically displays the difference in performance
between disks.
- "AnalyzeDisk" Utility
This application gives various information about the disk's reading/writing
capabilites and features. From that benchmark we mainly get the "Service
Read/Write" results:
IPEAK's service time tests uses 25.000 random, single sector read/writes at
all the disk's capacity. The resultant distribution of service times is plotted
as a probability density vs. service time. This curve represents the probability
that a request will be completed in a given amount of time. The service time
distribution is not cumulative. The test results can be directly compared to
the access time test. User can download the test results and through check the
various results himself.
User should be aware for any outliers: "...A well behaved storage
subsystems should have any outliers in the service time distribution and the
distribution should be narrow. Outliers in the distribution makes the storage
subsystem unsuitable for real-time applications, such as multi-media, where
variations in service time are detrimental..."
- "RankDisk" Utility
We are using Intel's default RAW traces for both DeskTop/Server workloads:
The height of each bar is the measured mean service time per disk access for
a drive stimulated with I/Os from the given workload. Shorter bars represent
better performance.
- DeskTop Trace Info
|
SysMark98 WIN98 (64MB)
|
SysMark98 WIN98 (128MB)
|
SysMark NT4 (64MB)
|
SysMark NT4 (128MB)
|
WS99 WIN98 (64MB)
|
WS99 WIN98 (128MB)
|
WS99 biz NT4 (64MB)
|
WS99 biz NT4 (128MB)
|
WS99 he NT4 (64MB)
|
WS99 he NT4 (128MB)
|
Read
|
18925
|
2838
|
2517
|
1522
|
18589
|
10157
|
9021
|
6238
|
19806
|
10915
|
Writes
|
7140
|
7946
|
1820
|
1534
|
4495
|
3898
|
4343
|
3700
|
21633
|
18653
|
Data Read
|
94.342
|
19.962
|
46.159
|
27.971
|
86.709
|
50.501
|
147113
|
113.814
|
396.050
|
196.420
|
Data Written
|
100.122
|
112.727
|
36.706
|
28.746
|
40.664
|
27.641
|
41874
|
30.191
|
847.036
|
685.423
|
Intel's DeskTop traces include mostly reads but also writes I/O operations.
- Server Trace Info
|
AIM3 w/ 5 Users (USL 4,2, 16MB)
|
SDET w/ 10 Users (USL 4,2, 16MB)
|
TPC-B Data Drive (USL 4,2, 192MB)
|
TPC-C Log Drive (NT4,0, 1GB)
|
Netbench w/ 5 Clients (NW 4,01, 32MB)
|
BYTE1 Unix F/S (USL 4,2, 16MB)
|
Read
|
126
|
8642
|
18243
|
7063
|
28792
|
1
|
Writes
|
40866
|
24423
|
18196
|
10365
|
31618
|
2244
|
Data Read
|
1.028
|
23.125
|
37337
|
14.465
|
117.932
|
0.004
|
Data Written
|
327.545
|
73.054
|
37265
|
35.680
|
110.051
|
9.107
|
At the Server strain mode, almost all traces include low reading and mostly
write I/O operations.
5. Heat
Test Methodlogy - Page 5
- Heat
To measure the temperature we are using the DTemp
utility. This utility through the IDE interface and the each drive's build-in
SMART functions can report the working temperature of each drive. The temperature
is taken after IPEAK's RankDisk test results as an general average result.