New LSI 9200 series controllers: 6Gb/s, 2.88 GB/s seq. reads, 1.87 GB/s seq. writes

[Anvil]

The file size doesn't necessarily have anything to do with the block size.
You can be working with multi gigabyte files and still the application reads/writes small blocks, it depends on the application/type of data.

Anyways, the 9260 is the better all-round controller.

[One_Hertz]

One Hertz! nice to see you, seems like it's been a while!

Well I concluded that no amount of extra storage speed will really reduce my game load times (which are the only thing I care about) so I lost interest in storage.

The file size doesn't necessarily have anything to do with the block size.
You can be working with multi gigabyte files and still the application reads/writes small blocks, it depends on the application/type of data.

Anyways, the 9260 is the better all-round controller.

A question comes to mind - can windows be forced to use smaller blocks when copying data? I think I read somewhere if you are copying a large file (a few GB) it uses 4MB blocks. Can that be changed to 128kb? It would speed up the transfer rates of 9211 users by a lot.

[Anvil]

One Hertz,

Not that I know of. (definitely worth checking out)

There are other options for copying files like e.g. TeraCopy.
(TeraCopy uses dynamically adjusted buffers, might not be optimal for the 9211 though)

Also, creating a command line file copy utility with a default block size of 128KB shouldn't be a problem at all.

[F@32]

Well I concluded that no amount of extra storage speed will really reduce my game load times (which are the only thing I care about) so I lost interest in storage.

At what number of drives you hit the plateau? I guess CPU/GPU speed and RAM becomes limiting factors...
Game loading is also the only thing I care about. Benchmarks are mostly to ensure I squeeze the last drop of performance from hardware I have.

[One_Hertz]

At what number of drives you hit the plateau? I guess CPU/GPU speed and RAM becomes limiting factors...
Game loading is also the only thing I care about. Benchmarks are mostly to ensure I squeeze the last drop of performance from hardware I have.

Single (good) SSD is as fast as it gets for storage. Stacking them in R0 does nothing for game loading times as CPU is the bottleneck in most cases. Some systems are also just faster for no apparent reason, i.e. Napalm.

[lowfat]

Well I concluded that no amount of extra storage speed will really reduce my game load times (which are the only thing I care about) so I lost interest in storage.

Feel the same way here

I'll still stop in the Storage section multiple times a day but I really couldn't give two effs about benchmarks anymore. I ran the Iometer script that GullLars created and I didn't even look or care about the results.

[Computurd]

everyone is entitled to opinions of course
i think hardware raid may not give you enough of a boost to warrant it for some, but ich10r is real good.

I dont know any single ssd that can load levels of L4D in 3 sec range or crysis in sub 8 seconds on initial loads.

so for me my array is worth it, but there is most definitely an very large area of diminishing returns that is reached quickly with raid. hardware raid cards aren't needed for most guys, especially if all they are looking for is game loads. But for professionals, benchers, and speed freaks, etc they are definitely worth it.

but i am running like serious clocks (4.6) and fast gpu's and uber fast ram. so you need a whole system approach to reap the rewards basically imo.

[Mastakilla]

just fyi:

A few months ago I asked in this thread how to see SMART values of HDs attached to this controller and no one was able to answer me

Awhile ago I discovered that the new version of HD Sentinel (3.20 pro) is able to see my HDs through the LSI RAID controller.
I can now see all SMART values of my HDs, including temp

[Computurd]

awesome, now that begs the question, if they can see the drives individually, why cant they issue the trim commands?

[bivaughn]

Hi all! Long (loooong) time reader, first time poster. After reading this thread and when building my new system, I decided to go with 9211-8i and 8x C300 64GB RAID 0 for my system + gaming volume. Unfortunately, after I got everything set up and booted into windows, my performance was not what I was hoping:

-----------------------------------------------------------------------
CrystalDiskMark 3.0 x64 (C) 2007-2010 hiyohiyo
Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

Sequential Read : 514.008 MB/s
Sequential Write : 529.183 MB/s
Random Read 512KB : 708.025 MB/s
Random Write 512KB : 545.441 MB/s
Random Read 4KB (QD=1) : 32.184 MB/s [ 7857.5 IOPS]
Random Write 4KB (QD=1) : 73.934 MB/s [ 18050.3 IOPS]
Random Read 4KB (QD=32) : 330.067 MB/s [ 80582.8 IOPS]
Random Write 4KB (QD=32) : 299.993 MB/s [ 73240.5 IOPS]

Test : 1000 MB [C: 6.2% (29.2/469.3 GB)] (x5)
Date : 2010/10/21 0:00:25
OS : Windows 7 [6.1 Build 7600] (x64)

Now, my original motherboard was an EVGA P55 FTW, and I was concerned that maybe the motherboard was not giving the 9211 enough PCIe lanes to do the job right -- returned the P55 and got a Rampage III Formula instead. As of last night I got everything reinstalled and witnessed identical performance.

The 9211 has the latest firmware and Boot ROM and I am using the latest Windows 7 drivers for the card -- I had similar issues with the stock firmware but noticed my IOPS go up significantly after updating firmware.

I'm kind of at a loss as to why 8 drives with quoted 355MB/s sequential reads can't even beat 64MB/s, and also why my random read performance is greater than my sequential read performance!

My next step is to isolate my drives (first I will try two groups of four) to try and see if one of the drives is broken and thus causing my slowdown. Do you guys think there is anything obvious that I am missing here, or anything I should be trying?

EDIT: It looks like Kristoferr experienced the same issue I did, although he only had four disks where I have 8. Should I really just resign myself to these abysmal speeds? Bummer, if so.

[Nizzen]

Looks like Areca 1880 is the way to go with C300

[Anvil]

@bigvaughn

The 9211 is a great HBA but it has it's weak sides.

If you download and run ATTO you'll see that the throughput is there, it's just very narrow. (128KB-256KB iirc)

Your 4K results are where they should be. (or very close)

If you need to run all 8 drives in one single array I would suggest moving up to a raid controller, e.g. 9260-8i or the Areca 1880.

What you can do test the performance/throughput on the 9211 is to create a software raid, it works great, it's not bootable though so you'll have to boot off one of the C300's or some other drive.

[bivaughn]

@bigvaughn

The 9211 is a great HBA but it has it's weak sides.

If you download and run ATTO you'll see that the throughput is there, it's just very narrow. (128KB-256KB iirc)

Your 4K results are where they should be. (or very close)

If you need to run all 8 drives in one single array I would suggest moving up to a raid controller, e.g. 9260-8i or the Areca 1880.

What you can do test the performance/throughput on the 9211 is to create a software raid, it works great, it's not bootable though so you'll have to boot off one of the C300's or some other drive.

Anvil:

Thanks for the response. CT posted in the OCZ forums a while back that the issue is more of a benchmark-related one than a performance in the real world (in this case, WoW loading times) one. Would you agree? I'm not opposed to purchasing a 9260-8i but if all I am getting is some benchmark improvements I am inclined to pass until the next generation of cards.

Thanks much,
Ben

[Anvil]

I agree with CT for the most part.

That is, for sequential IO, throughput has a narrow sweet spot, for most other tasks it rocks

[Computurd]

dude for wow loading times you wont find anything quicker than that, it will be a screamer!
is there a NF200 on your board? gah if it is.
might run a winsat disk and give us a looksie here....

elevate a cmd prompt to admin status
type in :
winsat disk

then give us a look please

the 1880 is a nice controller, but if i were to buy one today i would wait for the LSI 2208 series that is coming soon very soon....dual ROC @ 1.4 capable of 400,000 IOPS @ 4k random. should equal huge tons of fun for us with the low QD random with that kind of power.
umm what else...do you have everest? if you could do the everest/AIDA64 (either ones) disk benchmark we can see the latency...it is the 'average read access"

also, what of a AS SSD screenie?

[bivaughn]

It's a straight X58 board with no NF200. I have the card in slot 3, which is supposedly X8.

Here's the screens. Any tips on improving performance? 1.67GB/s seems abysmal.

[Computurd]

well winsat is a little hard to trick. It is saying you are getting 2.2 gb/s...maxing out that controller. your random of 565 is just uber! that is awesome
dont worry about benchies man.how does it feel? i bet its quick as hell.
you have to take into consideration that the 9260 maxes out at 2.8, but it is clocked higher with the saem ROC @ 800 mhz. for this 9211 with the roc @ 533 mhz that is damn good.
unless you are hellfired for setting world records and such there is no need for anything more. that is awesome. you have to realize stevero put up a top ten PCMV score (at one time, now its 12th) with a 9211...these things are killer. you could set a record with it actually

if you hunger for more, boot to one C300 (or two in raid 0) and run the 9211 as dynamic disk. you will see some amazing numbers, but unless you are clocked extremely high, it wont benefit you. for WoW i dont think even higher clocks will help.

TBH for WoW this is like robbing a convenience store with a missile launcher. (saw that in someones sig, but it works)

[bivaughn]

well winsat is a little hard to trick. It is saying you are getting 2.2 gb/s...maxing out that controller. your random of 565 is just uber! that is awesome
dont worry about benchies man.how does it feel? i bet its quick as hell.
you have to take into consideration that the 9260 maxes out at 2.8, but it is clocked higher with the saem ROC @ 800 mhz. for this 9211 with the roc @ 533 mhz that is damn good.
unless you are hellfired for setting world records and such there is no need for anything more. that is awesome. you have to realize stevero put up a top ten PCMV score (at one time, now its 12th) with a 9211...these things are killer. you could set a record with it actually

if you hunger for more, boot to one C300 (or two in raid 0) and run the 9211 as dynamic disk. you will see some amazing numbers, but unless you are clocked extremely high, it wont benefit you. for WoW i dont think even higher clocks will help.

TBH for WoW this is like robbing a convenience store with a missile launcher. (saw that in someones sig, but it works)

The performance doesn't feel like what I was hoping for, to be honest. My old array was a Highpoint 3510 with 4 of the original Samsung SATAII SLC 32GB units (so, 400MB/s reads) and it doesn't feel much faster than that. I haven't taken my system up to 4.0 yet (still hovering around 3.2) so that's on the list for tonight.

Thanks very much for the help in analyzing performance of the array -- hopefully my next controller purchase will be able to take more advantage of the raw read speed these units possess.

Have a great one, all!

[Nizzen]

dude for wow loading times you wont find anything quicker than that, it will be a screamer!
is there a NF200 on your board? gah if it is.
might run a winsat disk and give us a looksie here....

elevate a cmd prompt to admin status
type in :
winsat disk

then give us a look please

the 1880 is a nice controller, but if i were to buy one today i would wait for the LSI 2208 series that is coming soon very soon....dual ROC @ 1.4 capable of 400,000 IOPS @ 4k random. should equal huge tons of fun for us with the low QD random with that kind of power.
umm what else...do you have everest? if you could do the everest/AIDA64 (either ones) disk benchmark we can see the latency...it is the 'average read access"

also, what of a AS SSD screenie?

Winsat disk:

Areca 1880 4g + 1xCrucial c300 128gb on a Asus Supercomputer with 2xnf 200 !!


What harm does nf200 for me? I think it is not mutch more latency?

It is crazy, but I think one c300 on my 1880 4g is good enough

[bivaughn]

Winsat disk:

Areca 1880 4g + 1xCrucial c300 128gb on a Asus Supercomputer with 2xnf 200 !!


What harm does nf200 for me? I think it is not mutch more latency?

It is crazy, but I think one c300 on my 1880 4g is good enough

Wow, yeah that sucks. You have better responsiveness scores than me, even. :-(

Edit: I placed an order for a 9260-8i, should have it by Saturday. I'll post some Atto numbers with that.

[Nizzen]

Wow, yeah that sucks. You have better responsiveness scores than me, even. :-(

Edit: I placed an order for a 9260-8i, should have it by Saturday. I'll post some Atto numbers with that.

ATTO is usless with a controller with cache. It uses the cache too much.

IOmeter is the way to bench.

But winsat is fun also

Areca 1880ix-24 4GB

[odditory]

^ i'd say winsat is pretty useless w/ 4GB cache too. or was that the point

[trans am]

Guys I have a question about how number of CPU cores can affect LSI raid and general HD performance.

I am prepping my os and software of for pcmark 2005 onto a c300 with AHCI enabled using winvista 64.
I was still waiting for my single core amd 145 chips to come from newegg so in the meantime I used the only amd AM3 cpu I had available, a x3 740 3ghz cpu. I figured get in all installed and then when the 145 comes I can just drop it in and insulate and start benching. So I got the thing all set up and optimized with the x3 and ran some test HDD suite benches. What I did was boot off the c300 and then in the advanced tab in pcmark05 i would select the drive letter with the LSI 9260-4i with the 3 vertex2 50gb in raid0 using the auto config in the lsi bios setup. I was getting really good numbers and then I dropped the single core 145 in and I'm getting abysmal scores. I mean the hit is so bad Im starting to wonder If I needed to actually install windows with the single cpu and its the actual install that is bad or the single core cpu is bottlenecking the hdd performance that bad. If its the cpu bottleneck that is the culprit is there a way in the bios to allocate extra ram to the card to compensate for the loss or is it jsut up to me to run the cpu at a frequency that is so high that it will make up for it? Or should I reformat everything and do a fresh install with the single core cpu and see what happens?

heres a before and after:

[Computurd]

i personally think it is lack of cores, but only testing will tell for sure.

[trans am]

Yeah I agree and notice no l3 cache. is that going to be a problem also? I couldnt wait anymore for a response so I went ahead and reinstalling windows with that single core installed. Even on a single core intel 430 with no L3 cache doing the same thing I was hitting HDD - XP Startup 242.77 MB/s and HDD - General Usage 98.74 MB/s

I'll report back with the 2nd results and hopefully get to bench something useful tonight. THanks Paul. any recommendations you have for me will only contribute to the team as this contest is a team effort. I can only add that when I set write back in lsi MSM settings I was getting much better results in PCmark05 hd benches than our tried and true write through default settings with the FP key. Im just as surprised as you are.
edit...

I just got into windows on the new install and I can already feel the difference in speed. lemme do a real test for the record. hang on. I gotta install a bunch of crap so it might be a while.