Fastpath..

[Anvil]

GullLars,

The testing we did using my Vs (or was it the Ms) where adding workers scaled, do you remember the results?
(using the ICH iirc)

[mbreslin]

Possibly later this evening, just out of curiosity I'm going to try the kit cables. I was just running the hdd suite test again as you typed the pcmv question.

Up to 95k hdd suite score with the new fw.

[Anvil]

I don't think the cables will make any difference, It wouldn't hurt trying them though. (I'm using the kit cables, the only thing wrong with them is that they are a bit long)

95K is a great score, is that on the OS drives?

From my limited testing on the PCMV Suite, the SF drives are the ones that gives the best scores. (I've tried 2R0 using the LEs, X25-Es, G2s and the C300)
I've ordered a new GPU so I might continue the persuit next weekend.

[mbreslin]

I'm benching from an x25-v right now, I'm sure the drive score will go down a little when the os is on it but I've never gotten this high of a score even when benching an empty array like now. I'm going to give pcmv one more shot and then go back to permanent 24/7 setup.

[GullLars]

Actually, if you look at the IOPS/Average accesstime graph, he gets best scaling in excatly the area QD 8-32. Not only is it higher IOPS, but it's also at lower accesstimes.

Here are some numbers from SteveRo. I'll make a computurd vs tiltevros vs mbreslin vs stevero graph for 1W (and 2W for tilt).



Steve's numbers show no scaling with workers.
Steve, would it be very much hazzle for you to do a pass-through software RAID on your 1231ML with (some, or all of) your acards?

[GullLars]

Anvil, i have the numbers for the V's and M's somewhere, i'll make a new bench thread on this forum where we can do more stuff, for now we can keep this thread for fastpath, and a reference from Steve's acards.

[mbreslin]

100k pcmv hdd suite score now. Still tweaking.

[GullLars]

Here comes a big one: Fastpath arrays comparison


As expected, Steve's Acards own the low QD space, but already at QD 8 they cross with mbreslins fastpath C300 array. And at QD 32, mbreslins fastpath C300 array gets overtaken by tiltevros' 8R0 x25-M fastpath array.

If you have a look at the second picture, it paints a nice image of what loads the arrays handle best, and how well they handle them.

Man, i'd love to see 8R0 acards at 9260 with FP

[GullLars]

If you guys are up for it, i've made 2 new configs for you to chart a more detailed performance scaling at lower QDs. 1 worker QD 1-32, linear stepping of 1 QD. 4 workers, QD 4-128, steps of 1 QD pr worker = 4 total.
This makes for 32 runs of 30s = 16 mins.
Don't worry about the size of the datasets, i've become better at filtering the results

EDIT: there was a typo in the first "detailed" config i posted, i fixed it and uploaded the new one here:
4KB RR project detailed.zip

[Computurd]

@mbreslin---you might have missed it in the past but tilt is under NDA with lsi...thats how you explain it bro. he has secret sauce

settings and systems configuration from LSI direct (phone)

when he said that he wasnt kidding he is a lucky fella

[SteveRo]

if i get tired of waiting for the 1880 i should give fastpath/8xr0acard a try?

[Computurd]

yes yes yes!!!!

it would positively own... i knew you wouldn't be able to resist long...

[SteveRo]

when are we suppose to see a FP for the 9211?

[Computurd]

i am not sure, but someone i know might... i will buy back the 9211 if it convinces you to get the 9260...the 9260 will be better regardless, and i will need a controller for my soon to be retired array anyway, gonna refresh drives soon...

[Computurd]

ok gullars what do i run cache or non cache on the 9260.. i am going to try some different stripe sizes to find best access time, havent really tested for best usage yet, just for pcmark. but now on to more relevant testing...

[Tiltevros]

9211 wont have the fast path BUT BUT the intel version will.
Now for my results... i will say it again ONLY INTELS are OFFICIAL SUPPORTED ON LSI 9260FP

[GullLars]

@computurd: I've made a custom config for you for this speciffic purpose computurd.
It's 4KB RR QD 1-128 exp2 for 1 and 4 workers. You'll probably only need the 1 worker version.
The 1 worker version should take about 3,5-4 minutes to run.
computurd accesstime thingy.zip
You should be able to read the results yourself directly from the result file using open office calc (the excel alternative) using , (comma) as separator (the default suggestion). If you want me to organize the results in one excell document later, run all the tests you want and post a zip with the csv files with the names you want reported back, like {stripe}_{cache config}_{read ahead} etc.

EDIT: @tiltevros: i didn't know that. Interresting to hear they only "officially" support intel.

[mbreslin]

@GullLars: Will you make a graph if I submit a bunch of your 4k config results with different Ndrives? And if so I was thinking of just 1w/6w/12w or do you think the whole set is useful?

Nice of LSI to have a sas/sata 6gb/s controller that doesn't officially support the only 6gb/s ssds. I hope the 1880 smashes it and I can leave lsi behind.

[Computurd]

well gullars i have office 2010 so that is no prob with teh .csv files. i have technet is that the only difference in your file there?

[GullLars]

I would love it if you could run through the configs in this zip first for your current array: http://www.xtremesystems.org/forums/...8&d=1275095315
It's the "detailed" config posted above. Basically 1QD stepping from 1-32 for 1 worker, and 4-128 in 4QD steppings for 4 workers.
I'll make graphs for pretty much anything you want, as long as it's labeled so i understand what it is. I perfer it if you use 1 result file pr array setup, i.e. 8 result files for 1-8 drives, with 1/6/12 workers runs in each result file.

I would also love it if you could do a run in pass-through software RAID. The same goes for tiltevros and steve, if splitting the array isn't too much trouble. It's usefull to have numbers for 9260 without vs with fastpath vs pass-through softraid. I'm also interrested in if steve's 1231 will be have an IOPS limit around 60K for passthrough softraid.

EDIT, @computurd: i boiled the config down for you so it's less to run and more usefull output, since you won't ever see QD >128 in practice (i would guess). I use open office to read the result file and filter columns, and paste it in excell 2007 and filter rows there using the sort function.

[mbreslin]

By pass through I assume you mean putting each drive in their own R0 array and then making a dynamic stripped disk in windows? I just did that and tried the 8 worker config and got the same 160k at 8qd or so. Since I can't boot it it's not useful to me but I'm setting it up for pcmv hdd suite just to see.

[GullLars]

I was thinking having 8 drives as non-member drives from the controller and run a windows software RAID-0 on them, since that would bypass the ROC. 8R0+0 would go through the ROC, if i understand it correctly, and would still give you the same bottleneck.

[Tiltevros]

first of all.....

LSI 9260 is a 6GB SAS controller... C300 are not SAS drives... there is no any controller exept onboard or expansion cards( like ASUS) for SATAIII or SATA6GB.
so plz stop complaning and start realising the difference in companies profiles...

LSI is supporting INTEL because of the X25-E and after a year they starting supporting X25-m's whitch is SATA2.

Now if u want ARC its ok its ur choose but plz stop complaning about not supporting iteams...

Now for my Results... Im using Server enquipment whitch is best enviroment for Raid Controllers.
The Fast Path technology is tested in Cebit whitch means that u can confirm it by communicate with LSI that they
TEST THE FAST PATH WITH ONLY 5 DRIVES TO GET THE 150.000 IOPs 4k random.
Now im sharing every information that i can without getting in lines of my NDA so please respect that.

[Computurd]

you cannot do a dynamic array from the 9260 to my knowledge

[GullLars]

oh, ok
I'm off to bed now, it's 04.30 here local time. I'll be back on in 8-14 hours, if you post results in the meantime, i'll get to them while i eat breakfast/lunch/afternoon food, or whatever you want to call it :P