(SSDs) ICH9 vs Adaptec 5405 - anyone interested?

[NeedMoMegaHurtZ]

Thanks for sharing the testing results!

I am doing everything 10-15 times with reboots

Wow. You are very dedicated & thorough!

PS. If you have the time, I would like to see some results with the array filled (e.g. 80%-90% full of random files,etc...) to see the effects. I've seen some results where the performance hit is massive, I think many NAND SSD's exhibit this behaviour so far. Maybe some drives have allocated extra space for the swapping/garbage collection when OS visible capacity is near full ?

Cheers!

[One_Hertz]

Thanks for sharing the testing results!



Wow. You are very dedicated & thorough!

PS. If you have the time, I would like to see some results with the array filled (e.g. 80%-90% full of random files,etc...) to see the effects. I've seen some results where the performance hit is massive, I think many NAND SSD's exhibit this behaviour so far. Maybe some drives have allocated extra space for the swapping/garbage collection when OS visible capacity is near full ?

Cheers!

I am aware of that so all tests were done with the array filled up to the max with the iometer test file.

[NeedMoMegaHurtZ]

I am aware of that so all tests were done with the array filled up to the max with the iometer test file.

Good to know!

So was there any performance increase then when the storage was near-empty? at least on the ICH9R?

Obviously the 5405 dramatically improves performance, but I'm wondering how this affects the analysis at different filled capacity points? Does the controller/cache compensate for the simultaneous mixed read/write poor performance when at full capacity?

Nice results & setup eitherway,

Cheers

[One_Hertz]

Good to know!

So was there any performance increase then when the storage was near-empty? at least on the ICH9R?

Obviously the 5405 dramatically improves performance, but I'm wondering how this affects the analysis at different filled capacity points? Does the controller/cache compensate for the simultaneous mixed read/write poor performance when at full capacity?

Nice results & setup eitherway,

Cheers

To properly do that, you need to send the security erase command to the drives. Once anything has been written, even if you erase it, the performance will stay the same. I have the ability to send secure erase, but of course when I had the ICH9 that was not possible unless I was going to do an extra reinstall of everything, which I was too lazy to do.

[XS Janus]

"to lazy" would not be the word I would use in any context with anyone who has posted what you just have.

[WRC]

can you also run hdtach or hdtunereally quick?. I would like to see how if there is any difference

[One_Hertz]

can you also run hdtach or hdtunereally quick?. I would like to see how if there is any difference

I ran hdtach and there wasn't any difference in read speeds (both setups were 288mb/s, but the 5405 adaptec card had 0.3ms response while the ICH9 had 0.2, which explains the read IOPS drop).

[alfaunits]

I ran through the thread again but didn't notice - did you make sure the Intel setup had Write caching turned on? It seems Adaptec helped exactly with tests that do writing, and obviously this would make a difference.

[RealTelstar]

Great tests, thanks.
I'll wait the ones with the intel x-25 now ))

[One_Hertz]

I ran through the thread again but didn't notice - did you make sure the Intel setup had Write caching turned on? It seems Adaptec helped exactly with tests that do writing, and obviously this would make a difference.

Of course it was on

[swiftex]

interested in knowing comparison with Intel...........
when do you think you'll have new disks?

[flopper]

300 re-boots, 1min, 300min, 5 hours of re-boots not included time for application runs.

That is xtreme

Thx.

[jcool]

Very nice results - have you flashed the 5405 to the latest firmware? It specifically improves read performance

[Justifire]

What's the bottleneck for on-board controller?
The cache or the throughput on the pci-e raid controller ?

[jcool]

ICH10R is capped around 650MB/s throughput which is rarely a problem. It doesn't have dedicated cache like a hw raidcon does, meaning you'll still get the stuttering issue with Jmicron & MLC based drives if you run them on the ICH.

[[XC] Synthetickiller]

ICH10R is capped around 650MB/s throughput which is rarely a problem. It doesn't have dedicated cache like a hw raidcon does, meaning you'll still get the stuttering issue with Jmicron & MLC based drives if you run them on the ICH.

Does anyone know how much memory you'd need on the raidcard per drive? Say ____ for 2 drives, ___ for 4,6,8, etc? I know its probably not linear.

The Adaptec 5405 has 256mb ddr2 onboard. Would 128 suffice for 2 or 4 drives? That's the kind of question I am asking.

[One_Hertz]

Very nice results - have you flashed the 5405 to the latest firmware? It specifically improves read performance

Nope, didn't have the December one at the time of those benchies . For some reason its not letting me do the flash in windows at this time (the button is grayed out)... I'll get it done before I put the Intels in.

[jcool]

Flash in Windows.. are you crazy
I always flashed mine in DOS from a bootable USB key, works a treat.

@Synthetickiller: Hard to say really, more is always better ofc. But even 128MB will suffice to run a raided pair of MLC SSDs without encountering the stuttering issues and slowdowns.

[[XC] Synthetickiller]

Flash in Windows.. are you crazy
I always flashed mine in DOS from a bootable USB key, works a treat.

@Synthetickiller: Hard to say really, more is always better ofc. But even 128MB will suffice to run a raided pair of MLC SSDs without encountering the stuttering issues and slowdowns.

Thanks. Unless new controllers come out within the next year, I'll probably get a couple of 30 gig drives and a good raid card.

[Danne980]

Flash in Windows.. are you crazy
I always flashed mine in DOS from a bootable USB key, works a treat.

@Synthetickiller: Hard to say really, more is always better ofc. But even 128MB will suffice to run a raided pair of MLC SSDs without encountering the stuttering issues and slowdowns.

Isnt it possible to test >256MB files in atto? Say one 128MB and one 384Mb?

[swiftex]

Nope, didn't have the December one at the time of those benchies . For some reason its not letting me do the flash in windows at this time (the button is grayed out)... I'll get it done before I put the Intels in.

you probably don't have admin rights set-up on your card,
that's usually why you're not able to flash with ASM..........