X25-E strip size test results

[Ao1]

Set up: Adaptec 5405, X25-E x 2 in raid 0. Win 7, no tweaks.

The raid strip sizes were reconfigured online using Adaptec Storage Manager without rebooting or formatting the drives. The raid strip that the OS was installed on originally was configured at 128K, from here it was changed to 16K, then 1024K, then 512K then 256K and then back to 128K.

Reconfiguring the strip size online with the Adaptec Storage Manager seemed like an easy option to save time testing various strip sizes on freshly formatted drives, but may well have skewed the results in the process, albeit that would be an interesting result in itself.

The easy option actually ended up taking more time as a fresh install of Windows 7 takes around 10 minutes and reconfiguring the strip size took around 20 minutes. Next I may try the same exercise but with a reformatted drive for each strip size.

Before the first reconfiguration ATTO produced the following result with a 128K strip. (Didn't think to run CrystalMark)

[Ao1]

Results with a 16K strip

[Ao1]

Results with a 1024K strip

[Ao1]

Results with a 512K strip

[Ao1]

Results with a 256K strip

[Ao1]

Back to a 128K strip.

[alfaunits]

WEll, let me give you a hint - you wasted time with all these "tests". ATTO is the most worthless HD test around, even more useless than HDTach really

[Ao1]

Seems like the 256K strip is best, although I'm really kicking myself for not testing the original 128K strip with CrystalMark before I started reconfiguring the strip sizes.

[Ao1]

WEll, let me give you a hint - you wasted time with all these "tests". ATTO is the most worthless HD test around, even more useless than HDTach really

Thats why I used CrystalMark as well. What would you test with?

[n0shutz]

only show the first one next time lol =)

[Ao1]

only show the first one next time lol =)

I guess it depends on which benchmark programme you can believe.

Changing the array with HDD takes hours.

With the X25-E the best case was 8 minutes, worst case 37 minutes. The time got progressively worse, which makes me think that the drives were affected by the tests. Intel apparently asked reviewers to format the drives after each benchmark test because the drives adapt to usage patterns.

I chose to do the test this way for two reasons, one to find the best strip size and two to see what happened to the drive, which was never going to give a conclusive result either way, however I'm still inclined to think that a 256 strip is fastest.

These tests were undertaken on drives that have been used for two months following an OS install. The drives were 50% full during the test but that have been used at 95% i.e. 60GB of written data on 64GB array.

If nothing else it shows these drives can take a hammering and at no time have I noticed the slightest drop in perceived performance.

[spazoid]

WEll, let me give you a hint - you wasted time with all these "tests". ATTO is the most worthless HD test around, even more useless than HDTach really

I beg to differ. CrystalMark is the worst of the worst.




@ OP

Go download IOmeter.

[Ao1]

If certain benchmarks are useless maybe someone could explain why that is the case. Every benchmark out there has been slated by someone at one time or another, but no one is explaining why. It sounds like it should be a new thread, but hopefully someone can give an explanation as to why a benchmark is a waste of time.

[spazoid]

First off, both Atto and Crystalmark in this case basically just test the linear performance of the cache on your controller. Crystalmark does test random to some extend, but with IOmeter you can specify which usage patterns you wish to test down to very specific patterns. Mixing random writes with linear reads for example is a very effective way at testing if the setup can/will generate stutters.

[Marios]

Try this under Vista or Seven
Make a batch flie bnch.bat
Edit
Copy paste the following

winsat disk -ran -read -drive %1 -count 1 -iocount 256 -ransize 16384
winsat disk -ran -read -drive %1 -count 1 -iocount 256 -ransize 32768
winsat disk -ran -read -drive %1 -count 1 -iocount 256 -ransize 65536
winsat disk -ran -read -drive %1 -count 1 -iocount 256 -ransize 131072
winsat disk -ran -read -drive %1 -count 1 -iocount 256 -ransize 262144
winsat disk -ran -read -drive %1 -count 1 -iocount 256 -ransize 524288
winsat disk -ran -read -drive %1 -count 1 -iocount 256 -ransize 1048576
winsat disk -ran -write -drive %1 -count 1 -iocount 256 -ransize 16384
winsat disk -ran -write -drive %1 -count 1 -iocount 256 -ransize 32768
winsat disk -ran -write -drive %1 -count 1 -iocount 256 -ransize 65536
winsat disk -ran -write -drive %1 -count 1 -iocount 256 -ransize 131072
winsat disk -ran -write -drive %1 -count 1 -iocount 256 -ransize 262144
winsat disk -ran -write -drive %1 -count 1 -iocount 256 -ransize 524288
winsat disk -ran -write -drive %1 -count 1 -iocount 256 -ransize 1048576
winsat disk -seq -read -drive %1 -count 1 -iocount 256 -seqsize 65536
winsat disk -seq -read -drive %1 -count 1 -iocount 256 -seqsize 131072
winsat disk -seq -read -drive %1 -count 1 -iocount 256 -seqsize 262144
winsat disk -seq -read -drive %1 -count 1 -iocount 256 -seqsize 524288
winsat disk -seq -read -drive %1 -count 1 -iocount 256 -seqsize 1048576
winsat disk -seq -write -drive %1 -count 1 -iocount 256 -seqsize 65536
winsat disk -seq -write -drive %1 -count 1 -iocount 256 -seqsize 131072
winsat disk -seq -write -drive %1 -count 1 -iocount 256 -seqsize 262144
winsat disk -seq -write -drive %1 -count 1 -iocount 256 -seqsize 524288
winsat disk -seq -write -drive %1 -count 1 -iocount 256 -seqsize 1048576

Save it under c:\user\yourname\
Start run cmd and type "bnch c"
You may change "c" with any partition you want to check.
In case you have a small partition you get better results.
In case you have formated the whole disk as one partition you get reliable results for the whole disk or raid.
Though we may change the parameter -drive %1 with -n %1 (not applied in raid) to check the whole disk or raid %1 no matter how many partitions we have.
In this case we type bnch 0 or bnch 1 to check the first or the second disk respectively
The numbers at the end are IO sizes. I have used the full IO size range, but we may add any IO size we want to check as follows
-seqsize <65536-1048576> bytes
-ransize <16384-1048576> bytes
The iocount set at 256 produces quick results.
-iocount <256-5000>

In a cmd window you might also copy past the following to get results similar to HDTach.
winsat disk -seq -read -drive c -count 1 -iocount 256 -seqsize 65536 -v

[Ao1]

@ spazoid & Marios,

Thanks guys.

So if I understand correctly the benchmark results distorted by cache, which I guess is the same regardless if it is on a dedicated raid card or if it is built into the SSD itself (?) Does that mean that benchmarks are OK if there is no cache in the equation?

If the benchmarks are written for HDD and are accurate for HDD that use cache are those results also not affected in the same way?

@ Marios, thanks for this. I will try it out.

I'm using hard raid for two reasons, first for capacity and second because I find it is much more stable than onboard raid. All I'm really bothered about is finding the optimum strip size.

[alfaunits]

So if I understand correctly the benchmark results distorted by cache, which I guess is the same regardless if it is on a dedicated raid card or if it is built into the SSD itself (?) Does that mean that benchmarks are OK if there is no cache in the equation?

Right on both counts. However, if there is no cache in the test (on the controller, drive, volume level OS), you can't compare it to the test with cache.
ATTO in other words gives no useful numbers whatsoever (except for marketing tidbits).
HDTach at least gives you a number that has some (albeit very limited) meaning. At least in file copying it is useful - ATTO is not!

If the benchmarks are written for HDD and are accurate for HDD that use cache are those results also not affected in the same way?

They are affected - read again: ATTO is useless

[Ao1]

Thinking about this more I guess it was my mistake to benchmark without making it clear what I was trying to achieve.

My objective was not to find out how the X25-E performs. My objective was to find the best stripe size using 2 X25-E's in raid 0 with the 5405, so cache is an important factor to the equation.

That said are the benchmarks still useless and if so is IOmeter still the best for establishing this?

And I guess I should also add what I am using the drives for! General OS use, games & photoshop. I am not copying large files all day.

[alfaunits]

ATTO is ueeless any way you put it! (except marketing again)
ATTO numbers will not equate to anything in real-world, be it simple OS stuff or power use.

[Marios]

audienceofone
Please check the post with the bench file again, since I made a few corrections.
I think this is a reliable test to compare different stripe sizes so that you may get better results from your Intel X25-E SSD x 2 RAID-0.

[Ao1]

audienceofone
Please check the post with the bench file again, since I made a few corrections.
I think this is a reliable test to compare different stripe sizes so that you may get better results from your Intel X25-E SSD x 2 RAID-0.

Thanks Marios, I really appreciate the time you have spent to help me.

First result below with a 128K strip. Now I'm changing to a 256 strip to retest. Again I am changing the strip online with the Adaptec Storage Manager without reformatting.

Maybe I'm being thick but I'm still confused about why benchmarks are not valid. Why can't anyone seem to agree on a valid benchmark for SSD? I hear spazoid loud and clear, but I have also seem other posters (non marketing in origin) that have said the opposite.

[Ao1]

Results for a 256 Strip. I'm going to put them in an excel table to make comparison easier.

[Marios]

OMG those are GREAT results from two hard disks in RAID-0!!!
You got a 7.7 (128 strip) and 7.8 (256 strip) Disk Bench Rate on Seven!! Unbelievable ... LOL
Do you think it is because of your Adaptec controller?
Did you use one partition for both your disks or did you make more partitions?
I think it's time for me to get two of those diamonds.


You might also copy the results instead of making a snapshot.

Just select mark, then select what you want to copy and press enter
Then paste in notepad.

For Excell use three columns. The first is the following. Then add as many columns as you want with your results.
Then make a chart.


Random Read 16KB
Random Read 32KB
Random Read 64KB
Random Read 128KB
Random Read 256KB
Random Read 512KB
Random Read 1024KB
Random Write 16KB
Random Write 32KB
Random Write 64KB
Random Write 128KB
Random Write 256KB
Random Write 512KB
Random Write 1024KB
Sequencial Read 64KB
Sequencial Read 128KB
Sequencial Read 256KB
Sequencial Read 512KB
Sequencial Read 1024KB
Sequencial Write 64KB
Sequencial Write 128KB
Sequencial Write 256KB
Sequencial Write 512KB
Sequencial Write 1024KB

You may also try the long form with -v for the two most important results (measured by Seven) which are random read 16K and sequential read 64K.
This way you see results from all the areas of your disk and you may also make a graph like HDTach with excel.

[alfaunits]

I think the test is mostly ran in the cache, as those write speeds cannot be achieved otherwise even with 4 of those babies.

[Marios]

It is very difficult to reliably measure ssd disks.
I have one more idea. This is easy.
Could you please shrink your partition to the actual data size and run again the same test on drive C?
Something like a drive C 98% full with available disk space at about 2% should be OK.
Leave the rest of the raid-0 without any partitions.