LSI 9211-8i versus LSI 9260-8i versus Areca 1231ML-2G

[One_Hertz]

One Hertz,

Of cource, QD must be taken into consideration, but as a single (atomic) operation it is correct.

Have you tried to determine the actual QD during normal/heavy workload on your system?

How often do you do a single atomic operation

No I haven't found a good way to trace QD.

[Anvil]

One Hertz

Everytime I load or save a single file to/from my editor.
The point is that if the stripe size doesn't match your common file size one ends up not utilizing more than one SSD.
(one cannot tell which SSD a file is stored on and in theory all of your 10MB small files could be on one of the SSD's, using a small stripe size one is guaranteed that the file is distributed on more than one SSD)

[One_Hertz]

One Hertz

Everytime I load or save a single file to/from my editor.
The point is that if the stripe size doesn't match your common file size one ends up not utilizing more than one SSD.
(one cannot tell which SSD a file is stored on and in theory all of your 10MB small files could be on one of the SSD's, using a small stripe size one is guaranteed that the file is distributed on more than one SSD)

If you are working with a single small file then the time it will take to read/write it is already in the ms range. Say 32kb. The time to read it once the seek is complete is roughly 0.16ms on one 200mb/s SSD. Total with seek on a fast SSD is around 0.26ms. Who cares at this point? Don't forget that with smaller stripe sizes your seek time will rise because every SSD has to do the seek and the highest seek time will be the one that you wait for.

Having all 10MB small files fall on a single SSD in an array is basically impossible... And even then 10MB will be in the near instant range on a modern SSD.

[Jor3lBR]

Ok so here is Pcmark vantage HDD test for the LSI 9211-8i, Areca 1231ML-2G, LSI 9260-8i and intel ich10R (onboard raid).
The i7-965 and memory were set to 4.35 (132x33) and 1848 (8-8-8) woth pcie clock at 101.
The arrays used and (rather surpising) results for each test are as follows –

Ich10R – 6x acard 9010 R0 – 66175 pnts
9211-8i – 8xacardR0 – 60108
1231ML-2G – 8xacardR0 – 56210
9260-8i – 8xacardR0 – 47428
Ich10R – 1x Mtron Mobi 3000 (boot drive) – 12022.

I decided to chime in for the Areca team and just made my first PCMark Vantage HDD Suite run with 2x degraded Intel X25m's not the best needs improvements:

E8600 duo core + Areca + Intel X25m = 71151 points

Compare url:
http://service.futuremark.com/compare?pcmv=261161

[SteveRo]

I decided to chime in for the Areca team and just made my first PCMark Vantage HDD Suite run with 2x degraded Intel X25m's not the best needs improvements:

E8600 duo core + Areca + Intel X25m = 71151 points

Compare url:
http://service.futuremark.com/compare?pcmv=261161

Wow Jor, great score!
Is that with 12xx or 16xx controller and with just 2 drives?

[Jor3lBR]

Wow Jor, great score!
Is that with 12xx or 16xx controller and with just 2 drives?

Hey SteveRo, thanks I just upgraded my storage setup and it's been a pleasure so far.

It's the 1231ml + only 2 Intel's X25m. Blows my mind away.

[Computurd]

wow jor3lbr, 624 mb/s for windows media, thats 312 mb/s each drive. damn near the limit of sata2

[Nizzen]

wow jor3lbr, 624 mb/s for windows media, thats 312 mb/s each drive. damn near the limit of sata2

Cache is king

[Computurd]

if you look in the resource monitor, use the disk tab up top, then navigate down to the storage bar, windows will tell you queue depth. i am not sure how accurate it is, but hey the operating system is telling you so i dont know how much more accurate it can get funny thing is when you load games queue depth never goes above 1, or vary rarely. I have wondered before if queue depth is not going up because the system is so fast. maybe queue depth only goes up on slower systems during ceratian usage? maybe if the OS is waiting it stacks quqe depths? but if it is being 'served' fast enough it doesnt raise QD? however, when you load i.o meter and run queue depth at whatever file sizes windows shows the perfect amount of queue that the benchmark runs. so must be accurate.

[Jor3lBR]

Cache is king

The result cached is higher then the one I sent to Futurmark. Usually the second run. The first run after reboot is more accurate but still cache plays a roll on it. I am trying to compare apples to apples like SteveRo.

[SteveRo]

The result cached is higher then the one I sent to Futurmark. Usually the second run. The first run after reboot is more accurate but still cache plays a roll on it. I am trying to compare apples to apples like SteveRo.

What did you use as for an array blocksize?
I have heard that can make a big difference also.

[Computurd]

yes, and i dont see the problem with a cache run tbh...hell if you have it, use it!

[Jor3lBR]

wow jor3lbr, 624 mb/s for windows media, thats 312 mb/s each drive. damn near the limit of sata2

yes, and i dont see the problem with a cache run tbh...hell if you have it, use it!

Thanks Computurd! Here's my PCMark05 legit no cache or tweaks that are forbidden by HWBOT like the ones we saw on that other thread (Dis-consider the windows xp start-up test I had to do it while using the array to copy a 650Mb file at the same time the test where running to be under the 220Mb/s limit):

http://service.futuremark.com/compare?pcm05=2174341

What did you use as for an array blocksize?
I have heard that can make a big difference also.

I believe it was 128K but I have to double check that.

[Computurd]

Dr. Busch talks about raid performance with ssd's in the Q&A on this podcast @ 47min 20 seconds. It's a shame he does not elaborate more but clearly he is not impressed with hard or soft raid.

http://intelstudios.edgesuite.net/id..._MEMS002/f.htm

that is a very good link and has a lot of very good info thanks!
i did note however that he was mistaken when he replied, and he acknowledges that he was refering to raid 5, not raid 0, and they even go on to mention some stripe size recommendations for ssd and raid 0. you have to listen until 49:30 when he acknowledges his mistake and then they start talking about hte raid 0 again, and one of the guys even brags about his 500 mb/s on his raid 0 array at home alot of it is also that he is not referring to the concept or usage of raid itself, but the fact that the current (at the time) hardware raid controllers and software are not optimized for it with raid 5(or probably parity in general, he is talking about redundancy because he mentions that the image of the drive is a better backup system) i do not get video on that link, only audio , but i am going to listen to the whole thing later. captivating!

[Computurd]

when i was talking to titlervros in msn about the 9260 i had told him once about how i used to use the windows 2008 drivers before they released the win7 ones because they gave better performance.
NOW this whole time i have been trying to figure out what the is holding back my array and me and him talked about it a few times but it turns out he used the windows 2008 driver for the 9211 and that is what unlocked his sequential performance with 64k. and i will be goddamned cause i put the card back on, and sure enough i just went over 1300 with a degraded array. i hadn't been able to get over 1100 with a fresh array before. oh well looks like i am running the numbers again, as these better drivers may effect the game loads and definitely my little chart.
this goes to show the immaturity of this controller, it is so new the win7 driver needs major work.

[Computurd]

@jol3br- ya know it just hits me how good that pcm05 score really is, and how youve been floating around for the longest time trying to find the best solution for that specific application, i cant count how many times you've mentioned it! i am very happy to see you get a kick ass result at those HDD scores with pcm05. man, what a achievement! i see great things for you with this! keep up the great work your patience paid off

[Jor3lBR]

@jol3br- ya know it just hits me how good that pcm05 score really is, and how youve been floating around for the longest time trying to find the best solution for that specific application, i cant count how many times you've mentioned it! i am very happy to see you get a kick ass result at those HDD scores with pcm05. man, what a achievement! i see great things for you with this! keep up the great work your patience paid off

You are soooo right and after all those PM's and noob questions that you patiently answered.

Between you, SteveRo, Napalm, OneHertz and lowfat's posts I was able to figure out just the right combo for the setup I needed.

Thanks again for all your great feedback on these arrays.

[ehorn]

Dr. Busch talks about raid performance with ssd's in the Q&A on this podcast @ 47min 20 seconds. It's a shame he does not elaborate more but clearly he is not impressed with hard or soft raid.

http://intelstudios.edgesuite.net/id..._MEMS002/f.htm

Thanks for the link. Listening to Session 4 now.

[Computurd]

dude i was watching the live webcast of the sandy bridge presentation during IDF and they actually answered a question that i submitted in real time...even said my name! lol it was about if pci-e 3.0 will be implemented into the sandy bridge chip. dude i was so thrilled that they answered it RIGHT then. it was like the highlight of my geeky career as an overclocker god im a nerd!

@jol3br..man you list some big names there, lots of top guns on this site when it comes to storage! you are definitely one of them, and you are learning just like the rest of us, so never hesitate to ask anything! it is great to be on a site with this much experience and talent, even though we all rarely agree on anything

[Ao1]

@Computurd

Dr. Busch is a consultant that makes money from system engineering in context of putting parts together that deliver the optimum yield. He is more interested in putting a system together that works optimally. Intel on the other hand want to sell lots of ssd's. The Q&A seemed to get a bit awkward at that stage.

Will Akin talks about tiny strip sizes and a max of 64K. Yet smaller strip sizes with iometer do not benchmark as well.

I just don't get why raid 0 shows no real benefit for real apps in desktop use. Benchmarks show near perfect scaling but that is not reflected in real app use. Marios & Computurd have shown that significantly more writes are occurring on one drive within an array so how can the benchmarks show near perfect scaling? Then there is the added randomness of writes that occur in raid and raid overheads etc.

I'm hoping GullLars will shed some light when we (hopefully) get to see his graphs that show performance at different qd's.

[Tiltevros]

ok here is something to see... LSI 9211i8 windows 2008 drivers and 7 intels-m
there is a 20% off because of overhead i dont know how to make it lower but this is the best that my mind can think
the first on is 256k file seq. on 32outstandings IOS
the second one is 4k file seq 1 oustanding ios

PCIE latance at 89 (best result) and PCIE packet @ 128
this is the best result that i can take from the 5520 chipset

[trans am]

Well I finally got the damn thing working. Thanks to Computurd. I wasn't able to get the gigabyte UD5 to recognize the card. I'm guessing it only reads VGA cards for now. The board is still new so maybe with a bios update the engineers can make it work.

Here's a before and after Crystal disk bench.

Before (using 64kb stripe on AMD sb750)



AFTER (lsi 9260-4i) 64kb stripe

[Computurd]

Trans Am- dude that rocks i am totally stoked for you! dont thank me man your the one who put in the sweat and frustration and im sure a few choice words there as well! i am sure it was worth it once you seen them numbers! get yer OS and some games on there and enjoy! awesome!!
now we need to get our aussie buddy up and going...

[Computurd]

Will Akin talks about tiny strip sizes and a max of 64K. Yet smaller strip sizes with iometer do not benchmark as well.

you make a good point, and that of course goes back partially to the difference between the benches/real performance. there is that in-between somewhere inbetween the two. for instance, i have run a 16k stripe on the 9620-8i and my game load times increased tremendously.


on there i am doing some comparison to non-cache and cached runs of the games. nonetheless the point is the performance is there even though it doesn't look purty when you run some i/o meters, looks much worse than the 1 MB stripe. these times beat the 9211 load times. the stripe size on the 9211 is not changeable.
NOW consider that the 1231 only beat the 9211 by 2 percent in Vantage and the 9211 placed in the top 5 Hall Of Fame. that is power for sure, and vantage is a 'real' benchmark in the sense that it uses real apps and tasks as opposed to straight numbers. it actually runs video etc and measures the bandwidth, as opposed to i/o meter that just runs numbers in relation to a test file.
so now that i rambled, the point is that will atkins has a point, the tiny stripe sizes ARE faster!

so many different ways all of this goes, i thinks the only way to find the best setting is to test and do it with a targeted approach of a few games load times, then vantage HDD score and the low QD of each file size...find your perfect setting....
average user just get one disk

[trans am]

Trans Am- dude that rocks i am totally stoked for you! dont thank me man your the one who put in the sweat and frustration and im sure a few choice words there as well! i am sure it was worth it once you seen them numbers! get yer OS and some games on there and enjoy! awesome!!
now we need to get our aussie buddy up and going...

Yeah man we got to get him in the game! I was up late last night as you know and I got to the end of my rope. I basically did the following to get this rolling. after so many raid0 partitions on different controllers I had figured it would be best to wipe the drives. I have no experience with hdderase so I stuck with what I know. Killdisk. Using kill disk I wiped the drives and meanwhile I made a 20gb windows7 partition on a raptor and installed the updates and drivers for the chipset. then I rebooted plugged in the lsi and booted into windows and installed the drivers for the 92604i and then did the flash with you. once it flashed successfully and you went to bed. I plugged in my clean ssds to the card and set the disk settings for native ide. since this board has 2 x16 slots I had to go into that bios setting that shows vga order. it usually says PEG, PEG2 PEG3, PCI and so on. since peg stands for the 1st x16 slot closest to the cpu socket, peg2 is the next x16 slot and so on. I set it to the one with the raid card. then I set my apci to s1 and gen2. rebooted got into webbios made the array using the auto config method (it was late/i was drinking) rebooted and installed win7 dvd. plugged in the usb key with the drivers/it doesnt ask for them but i did it anyway. installed windows finally no reboots. installed the drivers/updates. turned off indexing/ power options to performance/ disk drive shutdown to NEVER. etc. etc. installed the lsi mega raid manager. went in and made the changes i didnt make when I did the automatic config in web bios. I wanted to try the 1mb stripe cuz that sounds awesome but I just wanted to get it going as fast as I could. 64kb is fine for what Im doing or for now. I was thinking we should make a LSI motherboard data base that shows which boards work and which dont and what you need to get them working.

heres a toast!

I've got some gaming to do.