Vertex LE vs Crucial C300

[onex]

It's 3 MILLIseconds, not 3 seconds.

this is quite confusing..
http://www.xtremesystems.org/forums/...&postcount=111

That the controller holds the files in cache for a little while (a few 100ms) and writes them as a bigger sequential block while remapping the LBA->physical table at the same time.

page 5,
you seem to have been confusing it too..

E:

apparently, if you take 0.003427 seconds (3.427 milliseconds) and multiply it by ~290 IOPs (1.19MBps/4096 bytes), you got 0.995 seconds..
so this is right lol ,
this is quite silly .

[Anvil]

I got delayed with a project so I've just started testing 2R0 C300 on the UD7

The 9128 controller wouldn't work using raid, not even using software raid.

I've started off using 4KB stripe size on the ICH10R.

Just a few teasers

cdm3_4kb_ss_wbcON.PNG

as-ssd-bench Volume0 30.04.2010 22-11-25_wbcOn.png

[SteveRo]

Wow Anvil, very nice! Anxious to see more - higher stripe sizes?

edit - 477GB capacity - being used to looking at acard capacities - this looks very nice also!

[Anvil]

Steve,

Capacity is perfect for my VM's.

I'll work my way up from 4KB up to 128KB on the ICH then I'll start testing on the LSI 9260.

[alfaunits]

Interesting how much ICH can cache
We'll need >12GB test files now

[GullLars]

Or just not activate WBC on ICH10R when testing...
I don't know if any of you use it on your ICH10R arrays?

[Anvil]

I normally don't use the ICH for raid and I've never used WBC for anything but testing. (on the ICH that is)
IRST 9.6 looks like a better performer using wbc so I might end up keeping it on for everyday use. (until I get a new raid controller at least)

Previously, using wbc performance dropped in some areas but 9.6 is almost identical using low stripe sizes whether wbc is on or off.
The exception is of course 4KB writes at low QD which gains a lot using wbc.

Here's CDM at 4KB stripe size

wbc off to the left, On to the right.
cdm3_4kb_ss_wbcOff.PNG cdm3_4kb_ss_wbcON.PNG

[Nizzen]

I normally don't use the ICH for raid and I've never used WBC for anything but testing. (on the ICH that is)
IRST 9.6 looks like a better performer using wbc so I might end up keeping it on for everyday use. (until I get a new raid controller at least)

Previously, using wbc performance dropped in some areas but 9.6 is almost identical using low stripe sizes whether wbc is on or off.
The exception is of course 4KB writes at low QD which gains a lot using wbc.

Here's CDM at 4KB stripe size

wbc off to the left, On to the right.
cdm3_4kb_ss_wbcOff.PNG cdm3_4kb_ss_wbcON.PNG

Looks like we have a winner combo here on ich10r

[turtletrax]

Newegg.ca had a stupid good deal on the C300 256GB right now. $629 shipping included!!

Newegg.com has them for $679 including shipping.

Damn. After Anvils results I want to grab another!! LOL

Again, thanks for all your work Anvil! Great stuff as always

[onex]

amazing scaling anvil,
read and write are almost the same,
huge write bandwidth at the 4KB,
yet curiosity rises as for QD of 4-10 on read 4KB,
if it goes as well as the 32QD, then Nizzen is right,
anyhow, seems like the ultimate dual RAID setup!

as for the earlier graphs lars showed,

and very strange results over 32KB vs the 64KB WB vs WT, seems as at high QD and larger block sizes, it's potential is being revealed,
it would be interesting to see a 1MB test with these.

it seems the WB cache is performing miraculously here,
have another look at the graphs and see how well the write through operates on the X25-M's at the small 4KB blocks vs the write back and how it changes at the bigger 64kb blocks ,
awkward results .

[mbreslin]

I never got around to trying 6 256gb c300s on ich. Anvil's scores kind of make me want to but I'm all setup now for 24/7 I don't really want to reinstall again. What if ich with 6 drives performed better than my 9260 with 8 drives? That would sure hurt.

[SteveRo]

I think 3x would probably max out ich10, 4 for sure.
4 would probably be the ideal number for the 1231ml4g

[SteveRo]

i think c300 beats acards in everything except 4k reads at low qdepth - this is 6xR0 at 128 stripe



Uploaded with ImageShack.us

[Anvil]

Looks like we have a winner combo here on ich10r

It certainly looks that way

Damn. After Anvils results I want to grab another!! LOL

That's what I'm thinking right now. 2 of these scream but what would 3 do?
I'll make up my mind when I'm done testing on the LSI 9260.

amazing scaling anvil,
read and write are almost the same, huge write bandwidth at the 4KB,
yet curiosity rises as for QD of 4-10 on read 4KB, if it goes as well as the 32QD, then Nizzen is right, anyhow, seems like the ultimate dual RAID setup!

as for the earlier graphs lars showed,

it seems the WB cache is performing miraculously here, have another look at the graphs and see how well the write through operates on the X25-M's at the small 4KB blocks vs the write back and how it changes at the bigger 64kb blocks , awkward results .

I've saved the iometer results for later, if GullLars want's to make the graphs he's welcomme.
I'll post the iometer result files anyway.

I never got around to trying 6 256gb c300s on ich. Anvil's scores kind of make me want to but I'm all setup now for 24/7 I don't really want to reinstall again. What if ich with 6 drives performed better than my 9260 with 8 drives? That would sure hurt.

You should have
3-4 of these on the ICH should be very interesting, it might saturate at 3, hard to tell.

I think 3x would probably max out ich10, 4 for sure.
4 would probably be the ideal number for the 1231ml4g

This is starting to get fun but also a bit expensive
I'll wait for the Areca 1880 reviews, if it performs like it should it moves to the top of my list.

I've finished 4-16KB stripe size testing, it looks like 16KB just might win.
I haven't had the time to study the iometer result files though.
(I'll post them later today)

edit:
I'm seriously considering buying an 890FX-SB850 board along with a Thuban. (6 core)
The MBs come with 6x 6Gb/s SATA ports , it could get to be a performance monster using the onboard controller.
Any thoughts?

[onex]

I'm seriously considering buying an 890FX-SB850 board along with a Thuban. (6 core)
The MBs come with 6x 6Gb/s SATA ports , it could get to be a performance monster using the onboard controller.
Any thoughts?

not sure about the SB850, the 790 seemed to saturate at about 600MBps,
yet you got this review at HardForum, with 4 1TB WD SATA3 drives going ~400MBps read, and show good scaling and stable results till the 3TB mark at ATTO, which is quite impressive,
no SSD RAID reviews so far aside from F@32 thread with 3 X25-M's G1?, showing seemingly low results for 4KB read at QD of 4 ~47MBps and the same for QD of 16..

E: still misunderstanding the 4KB random read non-scaling RAID behavior, the controller seems to give about the same results as others..

your dual 256GB C300 should probably saturate the controller.
i would go for Intel if the RAID is the weight when buying this setup, yet thuban is a nice chip to play with,
anyhow from what it seems, the SB850 should have some benefits over the 790 also from the RAIDing aspect, yet i would have second thoughts about buying it.

E: it should generally be fare enough,
Cecil havn't came back to us with updates on his dual SuperTalent 64GB RAID setup with this SB, so we might be missing something.

[F@32]

edit:
I'm seriously considering buying an 890FX-SB850 board along with a Thuban. (6 core)
The MBs come with 6x 6Gb/s SATA ports , it could get to be a performance monster using the onboard controller.
Any thoughts?

A bit slower than ICH10R as you can see from my preliminary testing here http://www.xtremesystems.org/forums/...d.php?t=246678
I will apply all the tweaks I learned about for SB850 and will properly test tonight/tomorrow.

[Anvil]

Thanks F@32

I'll check your thread for updates.

[Anvil]

I've finished testing using ICH and have just about started on a few tests on the LSI 9260.

iometer result files for C300 2R0 ICH wbc Off
C300_2R0_512B_256KB_ICH.zip

I'll post a selected assortment of benchmarks within a few hours.
Some of them are quite good

edit:
A one off run using 4R0 C300+LE (I might expand on this one later)
512B_256KB_RR_C300LE_4R0_4KBALIGNED_WBCON.zip

[GullLars]

I'm working on IOmeter data from Tiltevros and SteveRo now, i'll have a look at your afterwards Anvil

[Anvil]

Make him happy

I'll look into the graphs your'e creating for Tilt later.
It might convince me to buy the 9211.

BTW >140IOPS (2KB) using the last results file i posted

edit:
What graphs are you making for Steve?

[GullLars]

I just posted a graph for Tilt with 4KB random read QD 1-128, 1 vs 8 workers. 1 worker maxed out around 110K IOPS, 8 workers hit almost 200K with still some scaling from QD 64 to 128. He may pass 200-210K IOPS at QD 256 or with a more powerfull OC, as it seems somewhat CPU-bound.

EDIT: I will be making the same type of graphs as the 0,5-64KB QD 1-128 project. I think i'll add Tiltevros and Steve's numbers to the current lineup and post comparative graphs and the xlsx file.
EDIT2: steve has given me:
Stevero Frakenstien raid 0 ich10R 16k stripe 4x x25v 980x no oc no WBC Random Read 0,5-64kb exp2 QD1-128
stevero Frankenstien 4.8 ht off pcie 119 ich10 4x x25v 16k stripe wbc off Random Read 0,5-64kb exp2 QD1-128
stevero no oc 1231ml4g 4x x25v 16kstripe Random Read 0,5-64kb exp2 QD1-128 v4

[Anvil]

C300 2r0 256gb ich

The first one is 4KB wbcOn (forgot to enter the text)

[Tiltevros]

I just posted a graph for Tilt with 4KB random read QD 1-128, 1 vs 8 workers. 1 worker maxed out around 110K IOPS, 8 workers hit almost 200K with still some scaling from QD 64 to 128. He may pass 200-210K IOPS at QD 256 or with a more powerfull OC, as it seems somewhat CPU-bound.

i cant overclock but i can give u more threads
u got my results using only 8 threads instead of 16 that i have ;P

[GullLars]

You can simply add 8 more workers in the 8-worker config i gave you and run that. It will give QD 16-256. I can add that to the graph i posted in your "i'm back" thread
It's not much work compared to the 3D diagrams, since that data set is only like 8 runs compared to 80-100.

[SteveRo]

Anvil - they all look very good
Not easy to pick a winner in stripe size based on just CDM.