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

[alfaunits]

Hmm, from Tom's random file I/O tests, I don't find the card to be bad at all (either LSI card).
Do you realize 0.1ms results in >10000 IOps? And, Tom's charts show more than that.
http://media.bestofmicro.com/A/N/219...r_database.png
Now I don't see why you say the access time is correct rather

I don't see why people say this card is great... Sure the max sequential speeds are high, but the small file speeds aren't that good at all. The access times are also awful like from an Adaptec. What is so great? The areca 1231 is better but even on it the small file speeds and the access times are still not on par with ICH.

[One_Hertz]

Do you realize 0.1ms results in >10000 IOps?

It doesn't.

[NeedMoMegaHurtZ]

Please, can one of you guys with these cards post some longer file copy tests? or at least CrystalDiskMark 1000MB ?

Sequential bandwidth as measured by those benchmarks is not very useful if it's only good for <512MB files... I mean an average DVD is >4GB , a Bluray can be 25-50GB ...

Not trying to "diss" this card or anything , just want information for purchasing purposes

[F.E.A.R.]

Yes, FC-Test (create, read, copy)
Please try this 5in1-pattern. (ins - iso - mp3 - prog - win)

http://rapidshare.com/files/278280187/big.ptn.html

[alfaunits]

It doesn't.

0.1ms = 10 IO per ms, *1000 per second = 10000IOps.
?

[Rhys]

The thing is... faster IS faster whether it is easily noticeable or not.

Yes it is! no one can ever disagree with that.
Bottom line comes down to how much someone is willing to pay extra for almost nothing in return, dumb money will gladly pay more to receive zero tangible benefits.

On a side note, while important to allot of people, access time is not always the most important metric to the OP.

[One_Hertz]

0.1ms = 10 IO per ms, *1000 per second = 10000IOps.
?

With the que of 1... Those tests you linked use much higher than 1.

[F.E.A.R.]

0.1ms = 10 IO per ms, *1000 per second = 10000IOps.
?

You forget the filesize

[Biker]

Please, can one of you guys with these cards post some longer file copy tests? or at least CrystalDiskMark 1000MB ?

Sequential bandwidth as measured by those benchmarks is not very useful if it's only good for <512MB files... I mean an average DVD is >4GB , a Bluray can be 25-50GB ...

Not trying to "diss" this card or anything , just want information for purchasing purposes

Personally i'd wait for a PCI-e 2.0 8x spec SATA card...

While these SAS cards are good a dedicated SATA 2 / SATA 3 only card would be killer.

Lower latency and Xtreme bandwidth = WIN

Areca or LSI... who is going to be first out the door?

can't wait!

[lowfat]

Please, can one of you guys with these cards post some longer file copy tests? or at least CrystalDiskMark 1000MB ?

Sequential bandwidth as measured by those benchmarks is not very useful if it's only good for <512MB files... I mean an average DVD is >4GB , a Bluray can be 25-50GB ...

Not trying to "diss" this card or anything , just want information for purchasing purposes

What are you using for drives? Unless you are going over the limit of ICH*R there is no reason to go a RAID card as it won't benefit sequential read/writes.

[Biker]

What are you using for drives? Unless you are going over the limit of ICH*R there is no reason to go a RAID card as it won't benefit sequential read/writes.

If we haven't managed it yet I think everyone in this thread intends to in the very near future

[lowfat]

If we haven't managed it yet I think everyone in this thread intends to in the very near future

I meant the limit for the controller not the individual drive like the ~ 600MB/s that ICH10R can do. I know it is my bottleneck but the 9260-4i isn't available in Canada anywhere.

[Biker]

I meant the limit for the controller not the individual drive like the ~ 600MB/s that ICH10R can do. I know it is my bottleneck but the 9260-4i isn't available in Canada anywhere.

My ICH10R tops out at around 520mb/s. If I connect the same 2 drives to the LSi card I get 600mb/s read (2 x25m-G2 RAID 0 tested with AS SSD).

Edit:

ICH10R 2x X25-M G2 RAID 0 128k stripe:




LSI 9260-8i 2x X25-M G2 RAID 0 128k stripe:



The 4k results are pretty low on the LSI card but the sequential is significantly better.....

[alfaunits]

With the que of 1... Those tests you linked use much higher than 1.

So? The LSI card outperforms Adaptec in some of the random I/O, and by a very large margin even.
That just shows the latency is either wrong, or not important.
FYI, my 2xX25-E RAID0s on Areca have latency of 0.1ms in Everest.
But they sure as hell aren't slower than the X25-Ms in the graphs here

You forget the filesize

Which is related... how to random drive I/O???

[One_Hertz]

So? The LSI card outperforms Adaptec in some of the random I/O, and by a very large margin even.
That just shows the latency is either wrong, or not important.
FYI, my 2xX25-E RAID0s on Areca have latency of 0.1ms in Everest.
But they sure as hell aren't slower than the X25-Ms in the graphs here

Latency and the IOPs you keep talking about are two different things. Latency = IOPs only in linear access. Various apps use a combination of linear and non-linear access (lower que and higher que). Both mean something.

Biker - AS SSD can not benchmark raid arrays that have a good amount of cache hence the screwy results. Try IOMeter/Passmark/IOzone or anything that is actually a good benchmark and you will see the same ~500-520MB/s reads.

[alfaunits]

Latency and the IOPs you keep talking about are two different things. Latency = IOPs only in linear access. Various apps use a combination of linear and non-linear access (lower que and higher que). Both mean something.

I am talking about IOMeter database pattern I/O, which is random I/O.
I really don't get it what you are trying to say..

[One_Hertz]

I am talking about IOMeter database pattern I/O, which is random I/O.
I really don't get it what you are trying to say..

Let me give you an example - X25-E - 4KB 100% random reads. At que = 1 it is about 5.5k IOPS. At que = 32 it is 35k IOPS. The access time stayed the same but the IOPs went up 6 fold.

[alfaunits]

What I see is 0.2ms read latency here:
http://images.anandtech.com/graphs/i...5207/19507.png
Still donno why you consider this latency a problem...

[Biker]

Biker - AS SSD can not benchmark raid arrays that have a good amount of cache hence the screwy results. Try IOMeter/Passmark/IOzone or anything that is actually a good benchmark and you will see the same ~500-520MB/s reads.

Yep. I realise that AD SSD falls over with large cache (just look at the 4k results above) and it is therefore pretty limited in use.

Subjectively My system feels faster when I run my test OS array from the LSI card when compared to onboard...

I do not have 6-8 fast drives to hook up and then run all the benchmarks you guys are asking for (just not worth doing it with 2) but hopefully within a month I well get 8 SLC drives in for review then we can run all the numbers

Either way by then this might be irrelevant as some new SATA 2 / 3 only PCI-e 2.0 cards will be available

[F.E.A.R.]

Which is related... how to random drive I/O???

Sorry - my bad english. I don´t understand this question
Ask me differently (detailed).


@ Biker

What is the raw bandwidth (without cache - or sustained r/w)?

[NeedMoMegaHurtZ]

What are you using for drives? Unless you are going over the limit of ICH*R there is no reason to go a RAID card as it won't benefit sequential read/writes.

Yep I'm aware, thx for the reminder, I only have 1 x-25M G1, but I'm looking to get 4 G2's soon. I'm still on the fence about which card to get, hopefully Areca will bring some new ones to the table too

I am talking about IOMeter database pattern I/O, which is random I/O.
I really don't get it what you are trying to say..

I'll try to clarify:

The "Database" workload is 100% random, but 67% read, 33% write and 8k blocks only. Obviously you can't generalize that to other patterns e.g 4k transfersize, or 0.5k transfersizek etc... So for those specific workloads that fit this pattern, that test may be valid - but only for the queue depth. But you have to plot different queue depths, and measure IOPS. Yes , the equation Total IOPS = 1000ms / avg IO response time is true; but "IOPS" itself is useless without context or testing conditions. Similarly, "access time" is useless without clarifying what access time you actually are referring to. It's like saying I have 1000 pounds... Well 1000 pounds of what? coal or gold?

The "access times" read from such programs like hdtune , everest etc... only apply to a linear read or write sequential pattern. The equivalent pattern in IOMeter would be 0% random, 1 Outstanding I/O. This is not very useful, because in reality, different workloads may have very different patterns. For example, surfing web is supposed to be 20-40 queue depth or something like that. "access time" isn't a constant . It's a function of many different things including workload, firmware, CPU etc....So similarly "access time" is useless without context or clarification.

Moreover, "average" access times might not even be all that useful a measure - it's only 1 measure of many that you should look at (Important to look at the whole picture and context). Maximum response time or maximum latency might be more important - because this is perceptible to humans. For example, High maximum response times on 4kb small random writes at practically all queue depths was seen on the early SSD's with jmicron = stuttering.

Sorry for long post...

Anxiously awaiting more benchmarks...

[One_Hertz]

The "access times" read from such programs like hdtune , everest etc... only apply to a linear read or write sequential pattern. The equivalent pattern in IOMeter would be 0% random, 1 Outstanding I/O.

It is 100% random actually. It is the time it takes for your array to access a tiny piece of data at some random location.

[NeedMoMegaHurtZ]

It is 100% random actually. It is the time it takes for your array to access a tiny piece of data at some random location.

Sorry, I was referring to HDTune in benchmark mode (which is the value that most people quote with the pretty graph), not random access mode (which is obviously random). Are you saying this is 100% random?

[NapalmV5]

funny how no one posts real apps performance on these controllers lol

1231 and 4x strong slc ssds decimates these sas controllers @ real apps.. except @ transfer greater than 800mb/s

[NeedMoMegaHurtZ]

funny how no one posts real apps performance on these controllers lol

+1 !!!

But can't comment on which array is "better" for certain tasks, because no one has done the tests..

I would love to see some real apps and file transfers etc... Nothing can be real testing

But it is also time consuming, so understandable....

Cheers