|
Re: 4x 32GB OCZ SSD Core 2.5" SATAII solid state disks tests
|
17-08-2008, 05:44 PM
#2 (permalink)
|
4x32GB OCZ SSD Core Raid Preparation
I plan to try out raid for both WinXP Pro 32bit and Vista Ultimate SP1 64bit with on board motherboard raid controller as well as use 3rd party hardware raid controllers eventually as well.
To prepare for the on board motherboard raid controller usage, I had tried to integrate the drivers into a new nLite unattended OS build from WinXP Pro SP3 disk. But that failed to install properly, so will be rebuilding a new nLite OS build from WinXP Pro SP3 disk but not make it unattended so I can use F6 method to load the raid drivers if the integrated driver method doesn't work.
I will follow Fernado_1's guide
http://www.msfn.org/board/Integratio...e-t107504.html to try integrating the
Intel Storage Manager v8.5.0.1032 dated 08/03/2008 package.
EP45 Extreme
4x 32GB OCZ Core SSD raid 0 - 128K Stripe on ICH10R controller
The EP45 Extreme F6c bios came with Intel Storage Manger ROM version 8.0.0.1039 which i used to created 4x32GB OCZ Core SSD raid 0 array with recommended 128K stripe size. Here's what you see at boot up prior to windows loading. AHCI bios installed is also displayed - does this mean AHCI is enabled by default ?
Then I did a fresh WinXP Pro SP3 install (technet SP3 cd no-slipstream) which had integrated Intel Storage Manager 8.5.0.1032 raid drivers. The 128GB raid 0 array was partitioned as illustrated below:
Intel Storage Manger 8.5.0.1039 nLite integrated driver
Benchmark Apps used:- Attobench v2.34
- Everest Ultimate 4.50.1469 Read Test Suite
- HDtach 3.0.4.0
- HDtune 2.55
- PCmark05 HDD Tests - DEP disabled only
Notes straight from
OCZ config/setup PDF guide:
- HDTune uses fixed strides across the array to measure small block sizes. If these blocks fall on page or chip boundaries, the time delays result in atypically low “calculated” performance spot data that are not representative of the drive’s real performance but an artifact stemming from inadequate testing methodology.
- HDTach’s fixed strides between sample points causes similar issues as those outlined in HDTune, in this case, the periodicities of the stride pattern and the flash memory address space cause some wave-like interference patterns of the benchmark results that are also measuring artifacts rather than being indicative of performance.
- AHCI: the Advanced Host Controller Interface allows Hot-Plugging and Native Commands Queuing as well as multithreaded access of the drive by applications. Enabling AHCI results in conflicts between the controller and the drive that are apparent as sluggish overall system performance.
- Benchmarks that actually work are those that are not geared towards quick and dirty assessment of a HDDs performance based on optimization of the test algorithms to meet the typical HDD architecture. Examples are PCMark Vantage, Winbench 99 2.0 “Drive Inspection Test” or ATTO.
- Differences in Addressing Compared to HDD: Conventional hard disk drives utilize rotating platters that are divided into cylinders (tracks), where each track contains a certain number of sectors. Each sector is 512 Bytes long. In addition, HDD platters contain servo bursts that are reserved areas for track information etc. Most HDD benchmarks are configured such that the benchmark effectively tests the data area, if a test block is given that spans across a servo burst or else from one track to the next, then the benchmark usually generates inconsistent benchmark results.
- SSDs use a completely different address system, which is based on “chips,” “planes,” “blocks,” and “pages.” Each page contains 4 Kbytes of data, however, because of the parallelism at the back end of the controller, every access includes simultaneous opening of 16 pages for a total accessible data contingent of 64 Kbytes. Each block contains 128 pages and the number of blocks per chip defines the total density of the chip.
- PCmark Vantage - The HDD test is intended to mirror real life applications and is recommended as a key comparator versus other SSDs and HDD. This is currently the benchmark of Choice for SSDS or HDDs.
- Standard HDD benchmarking software - The industry standard benchmarks for the last decade have been Winbench99, especially “Drive Inspection Test,” IOMeter, and ATTO. For general media performance WB99 2.0 is still the test of choice since instead of spot checking, the entire storage media is tested. This causes the test to run longer than HDTach or HDTune but the results are accurate representations of the media performance across the drive instead of extrapolations of spot data.
- ATTO is the preferred Standard benchmarking software as while it runs spot data is has several advantages over other HDD benchmarks including the fact that It shows the specific speed of each file size transfer and shows a true average , this benchmarking product also doesn’t seem to favor SRAM over DRAM caches .
System:- E8500 Q740A479T 2L7 / 1313
- CPU Cooling: Dtek Fuzion Rad Tower Box
- Gigabyte GA-EP45 Extreme rev 1.0 F6c bios
- 128MB Gainward FX5200 PCI
- 2x1GB Crucial Ballistix PC2-8500 dual channel kit
- Memory cooling: 120x25mm Thermaltake 81cfm fan
- 4x 32GB OCZ Core SSD 2.5" solid state disks raid 0 - 128K stripe
- Pioneer 215 DVD-RW
- 1200W Silverstone OP1200
- WinXP Pro SP3 Nlite with integrated Intel Storage Manger 8.5.0.1032 raid drivers
Results:
As mentioned by
OCZ, each app resulted in varying results due to the way the app tested the SSD array. Attobench (which OCZ recommends as the app to use to test OCZ SSD Core series) and Everest probably showed a more accurate result. HDtach and HDtune showed 4x32GB OCZ Core SSD raid 0 results similar to that of a single non-raid 32GB OCZ Core SSD though which I'm sure is definitely wrong ?
I ran each app in 2 configurations:
Config 1: with DEP enabled by default with Performance Option set to default Programs
Config 2: with DEP disabled and Performance Option set to System cache.
Write back Hard drive data cache on raid 0 array was set to enabled
Results will be displayed with Config 1 (left), Config 2 (right).
Attobench v2.34
C drive - OS partition
E drive
F drive
Everest Ultimate 4.50.1469 Read Test Suite
425-493MB/s speeds with DEP disabled with System cache priority giving slightly faster performance.
HDtach 3.0.4.0
Virtually identical performance although results closer to non-raid single 32GB OCZ Core SSD and no where near what 4x32GB raid 0 config should be. Probably, hdtach needs an update for ICH10R / 4x32GB SSD raid 0 ?
HDtune 2.55
Similar to HDtach - showing non-raid single SSD like performance.
Update: Seems I might not have write back caching enabled for the raid volume itself which could explain weird results ? Only have hard drive data cache enabled for the array.
hard drive data cache enabled on array
Volume Write back cache disabled
After some reading, seems Volume Write back cache is an option for Raid 5 configs only.
PCmark05 HDD Tests - DEP disabled only
Installed Pcmark05 on the SSD raid 0 array - which took 17 seconds to install
I only did the PCmark05 HDD tests and since I have the Advance paid version I ran 3 tests to test each partition on the SSD raid 0 array - C, E and F drives. Highest score again was on the partition with the most free disk space - Drive F with a score of 12,667.
I plugged in my original non-raid single 74GB Western Digital Raptor 8MB drive which is full of data and OS etc, and chose the most free disk space partition - Drive J to run a PCmark05 HDD test for comparison.
Tabulated results:
Update: Oct, 2008 Wolf-tech has some interesting info posted in this thread
http://i4memory.com/98121-post31.html about a known issue where setting up raid in
non-consecutively plugged in SATA ports like I did above will result in lower performance similar to single disk performance. I didn't have a chance to test this theory out as the system has been broken down to make way for new boards to test. But you folks are free to test this out and report your findings here
4x 32GB OCZ SSD Core 2.5" SATAII solid state disks tests
