Anvil's Storage Utilities

[Anvil]

@Anvil- i would like the option to retain a static test file as well. This would be helpful. Even with the *now* well-known uber longevity of these drives, no need to speed along degradation in raid arrays.
-also one last wish 512B @ QD 128 so i can show 465,000 IOPS

doing some long-run tests this week for a article, but i will put the toys up and do some playing soon so i will post some results.

I've been too busy lately, I've already made the necessary adjustments to retain the test files and will send you that build later tonight.
Will need some feedback on that though as SF based drives are behaving "differently" from other drives, especially in this regard. (static test files)

[Computurd]

Anvil- your awesome man, i really appreciate your time, you are the guru of the guru's

[vxr]

how can I use the iometer???

[sanctified]

Great utility! Nice work.

[Anvil]

how can I use the iometer???

The IOmeter menu option is not what you think, it was initially an option for importing iometer result files
Having imported the result files one can export to Excel /compare results based on a lot of options.

The IOmeter "like" benchmarks in my App is found in the benchmarks menu (Threaded IO)

[Nizzen]

16gb testfile

[One_Hertz]

There you go, those are mostly real results. Except the 32k and 128k randoms show a lot higher than they really are for some reason. (real results for your array should be less than 228MB/s for 32k and less than 920MB/s for 128k.)

Anvil - any idea why? QD1 RR 32k and 128k read iops must be less than QD1 RR 4k iops by definition but his result is showing much more? Is it a bug in the software?

[Vapor]

Bit of a twist for this run....

Parallels VM of Win7 SP1 on top of OS X 10.7 with the VHD on an Intel 80GB G1, default settings across the board. Definitely looks like there's some caching going on



*Might* be some NTFS-like compression too, have had a hard time turning off that setting

EDIT: yeah, definitely some compression going on (especially on writes), re-ran with incompressible:

[Computurd]

8x 128 gb wildfire R0 9265-8i with FastPath. no cache

[SteveRo]

^^ wow, very nice Paul!

[Nizzen]

Nice Computurd, but try even larger like 16GB test

I have to try with more than 4x corsair gt. 120gb. This benchmarks is THE ONE

[NapalmV5]

lol @ no cache/cached statements.. mesmerizing

if you havent noticed every result posted in the thread is cached regardless of controller

no cache?? then physically remove the ssd cache/raid controller cache/ram/cpu cache/all system cache/all os cache/every bit of cache

so what you left with then ?? may as well throw your system out the window

[johnw]

Anvil:

Any chance of adding a test similar to the SNIA conditioning test? I'm thinking of something like:

1) User chooses random or sequential, IO block size, read/write mix, test file size (span), and test duration

2) Program runs with selected parameters while measuring IOPS, for the duration specified

3) Program produces a graph of IOPS vs. time

After that functionality is working, it would be cool to be able to define macros to repeat the test with different parameters (or cycle through all possible parameters)

[Computurd]

I concur. we definitely need a set of tests that could emulate what SNIA isdoing, in a manner that can be easily replicated by everyone. would be great, i know you are extremely busy though!

[Anvil]

I'll have a look at the SNIA document.

It is interesting and i have been playing with specific parameters that looks to rapidly degrade performance.
(specific block sizes, both on sequential and random I/O looks to do the trick)

[SteveRo]

Security just let us back into the bldg, i spent about 30 seconds under my desk before we all ran for the doors. 5.8 earthquake in central VA. My tummy is a little upset.

[Anvil]

Hope you are all OK!

[Computurd]

holy crap setve-o! i hope everything is fine and that more importantly that your house and loved ones are fine as well!

[trans am]

Holy crap! omg Paul!
@ Anvil I saw a thread about HWbot hdd category and I was wondering how you would go about making a version dedicated to HWbot that could be the end all be all hdd bench? What needs to happen? Why is this such a touchy subject nobody wants to step in?
I think Anvil has the best one so far. Why not elaborate on this? how the hell can this make it into hwbot without cheat accusations and BS? There must be some way to get this on the Bot.

@ steve I was on the 17th floor in my nyc office and I started feeling this dizzy feeling like I was tripping or this vertigo feeling. when the painting started wobbling I thought to myself "Ny is on a fault line and I think this was my 1st earthquake experience" sure enough. Man you were at the source!!! It was crazy!

just got 2 m4 64gb yesterday and added my c300 (3 drive 128k partition R0 lsi 9260-4i fpKey)
I never just benched the 2 m4 as a matched set so maybe I should. Or just get 4 m4. I just went to the 3 drive array so maybe the c300 is slowing me down. Or its the lsi on the x4 slot that is taxing me. anyway. I tried going with pauls 4gb method to eliminate the cache. this is write through / direct io and no read ahead. going write back and read ahead gave worse results. 128k stripe.

[SteveRo]

All is well on the home front, no breakage or damage to the house but lots of stuff moved around on tables and stuff that fell to the carpet. Very unsettling experience, I hope we don't have anymore

@ transam - yes, I was in on the 6th floor in our bldg - I hope we are done with earthquakes!

edit - home is 59 miles dues north of the epicenter, the epicenter was very near the Lake Anna Nuclear Power Plant

[Anvil]

@ Anvil I saw a thread about HWbot hdd category and I was wondering how you would go about making a version dedicated to HWbot that could be the end all be all hdd bench? What needs to happen? Why is this such a touchy subject nobody wants to step in?
I think Anvil has the best one so far. Why not elaborate on this? how the hell can this make it into hwbot without cheat accusations and BS? There must be some way to get this on the Bot.

I haven't had the time to contact hwbot but it is of course interesting.
Cheating is detectable!

[squick3n]

3x m4 128GB R0 settled state

I ran the test twice. Both times I got a "Not Responding" error right at the beginning, but the test started back up after about 10 seconds. Not sure what numbers I "should" be getting, so I don't know the effect of that brief pause.

Great tool otherwise. Amazing work Anvil.

Running Vertex 3 Max IOPS x2 on an Intel system. Much better. Wonder what was different.

[Boogerlad]

There you go, those are mostly real results. Except the 32k and 128k randoms show a lot higher than they really are for some reason. (real results for your array should be less than 228MB/s for 32k and less than 920MB/s for 128k.)

Anvil - any idea why? QD1 RR 32k and 128k read iops must be less than QD1 RR 4k iops by definition but his result is showing much more? Is it a bug in the software?

it is not a bug, you're just misinterpreting the results. mb/s is iops*transfer size.

[One_Hertz]

it is not a bug, you're just misinterpreting the results. mb/s is iops*transfer size.

Nope didn't misinterpret anything.

[Anvil]

I'll do a few tests using my Areca, didn't notice that comment until now

I can't remember seeing other setups resulting in anything like it, a bit weird.

Pauls review of the Wildfire shows no such effect on the 9265 using lots of drives. Link