LSI 9260-8i Versus Areca 1231ML-2G

Good morning all,

General Observations –

Controllers: Both the LSI 9260-8i and the Areca 1231ML-2G take about the same time to start up – seems like ~ 20-25 sec.
Both seem very reliable, I did not encounter any errors or dropped arrays, which is saying something because I have significantly stressed both controllers over the last two days.
Both controllers use the same cable connectors – I used both Highpoint and Areca (fan out) cables with no problem.
The size of the 9260 is an advantage in tight places and it seemed that the 9260 ran cooler than the 1231.
The 9260 has a couple advantages over the 1231 – pcie 2.0 bandwidth to the motherboard and price.
The 1231 transfers data to the motherboard at only pcie 1.0 and is $200-300. more expensive than the 9260.
The 1231 has a couple advantages over the 9260 – 2GB vs. 512MB onboard memory cache and the 1231 offers native SATA support.
The 9260 is both a SAS and SATA controller in one, perhaps SAS via h/w and SATA via s/w? Could be.

Drives - I use what SSDs I have on hand.
The SSDs I have are perhaps not representative of the average SSDs that most people are now using.
Obviously, your results will vary based on the drives you use.
The Acard ANS-9010 is a DRAM based SSD and the Mtron Mobi 3000 is an early generation SSD based on SLC memory.
Both are SATA (not SAS) drives.
From a random access speed perspective the acard and the mobi are at somewhat opposite ends of the performance spectrum.
The acard 9010 is extremely fast at random IO while the mtron mobi is extremely slow (4KB writes) - by today’s standards (about the same as a good platter drive).
In sequential read writes, both are on the slow side – the acards come in at ~ 175/150MBps read/write per drive while the mobi’s read/write at ~ 100/80MBps per
drive.

Test set up – Test system is i7 965 on a Gigabyte Extreme mobo, 3x1GB G. Skill 2000/C9, 2xGTX295 Quad SLI.
Boot drive via Mtron Mobi 16GB on the ich10r controller.
All arrays are standalone on the controller (only the array to be tested is attached).
Clocks - cpu @ 4.48 (35x128) ht off, memory @ 1792 (8-8-8-30), pcie @101

LSI 9260-8i set up – I tested several variations and found that Biker’s set up was the best – see – http://www.xtremesystems.org/forums/...&postcount=112
All 9260 arrays are set to 64-stripe size (optimum for the arrays used in this test).

Areca 1231ML-2G set up – After testing several variations I found that the default set up is the best.
All 1231ML arrays are set to 64-stripe size (again optimum for the arrays used in this test).

Acard array set up - Four acard 9010s, all set (via jumper) to raid 0.
So 8xR0, each drive is 4GB (2 sticks of 2GB each DDR2).

Mtron Mobi 3000 array set up – 4 16GB SSDs all in R0

Summary of performance – Well, for the SSDs that I can bring to the test, I don’t think there is a clear winner here; both controllers have their strengths and weaknesses.
The LSI 9260-8i has the best sequential read speed with the acard array and the best workstation IOps with the mobi array.
The Areca 1231ML-2G has the best 4KB random speed for both arrays, the best sequential speed for the mobi array and the best workstation IOps for the acard array.
See the iometer summary table below –

http://img12.imageshack.us/img12/554...l2gcompari.jpghttp://img12.imageshack.us/img12/926...jpg/1/w716.png

Several benchmarks follow for each array on each controller - interesting that HDTach doesn't seem to work on the 9260 for reads.

Benchmarks of Acard ANS-9010 8xR0 on LSI 9260-8i -

http://img40.imageshack.us/img40/180...iacard8xr0.jpghttp://img40.imageshack.us/img40/926...pg/1/w1762.png

Benchmarks of Acard ANS-9010 8xR0 on Areca 1231ML-2G -

http://img17.imageshack.us/img17/100...ml2g64stri.jpghttp://img17.imageshack.us/img17/8xa...pg/1/w1754.png

Benchmarks of Mtron Mobi 3000 16GB 4xR0 on LSI 9260-8i -

http://img182.imageshack.us/img182/6...064kstripf.jpghttp://img182.imageshack.us/img182/9...pg/1/w1810.png

Benchmarks of Mtron Mobi 3000 16GB 4xR0 on Areca 1231ML-2G -

http://img32.imageshack.us/img32/168...l4xmtronr0.jpghttp://img32.imageshack.us/img32/123...pg/1/w1768.png

Thanks for posting this and doing all those test SteveRo!

This is what we need: head-to-head testing

There was mention of some firmware revisions for the LSI in another thread, were you using the most recent?

So what is the plan for you with the controllers?

Cheers

Quote:

---

Originally Posted by NeedMoMegaHurtZ

Thanks for posting this and doing all those test SteveRo!

This is what we need: head-to-head testing

There was mention of some firmware revisions for the LSI in another thread, were you using the most recent?

So what is the plan for you with the controllers?

Cheers

---

I think I'm using the latest f/w for all controllers/drives.
The latest f/w for the 9260 is 57 as of yesterday this time.
Also I should add - all test where done under xp-32.

very nice! I don't suppose you did any appliation launching and/or photoshop?

Quote:

---

Originally Posted by lowfat

very nice! I don't suppose you did any appliation launching and/or photoshop?

---

I haven't taken the time to figure out the application launching thing yet.
I do have an old version of photoshop someplace but it would take some time to find it.

Thank you for taking the time to conduct these tests and post your results!

Great this is exactly what we need. Thanks!:)
It would be great If you could find the time and do some head to head app runs and software install times between mobi arrays on both controllers!
I guess that would pretty much settle the argument from the other thread.:)

1231 vs 9260
800mhz vs 800mhz
sata2 vs sata2
hard vs soft
@ real apps theres no comparison

come on steve.. some real apps.. i know what the outcome is.. but do it for the ones that need more convincing..

janus - i aint getting/spending another dime on sas controller.. just waste of money/time

Thanks for testing :)

As your results show the LSI card's 4k performance is hurt by it's SAS overhead (and possibly smaller cache) with higher latency meaning lower small file performance while the Areca falls behind at high sequential speeds due to it being PCI-e 1.0 and previous generation.

Informative results indeed :up:

SteveRo

Thanks for the comparisons.

Could you please list the parameters you used for the iometer tests?
I've got some Intel's I'd like to compare to your results using the LSI 9260-8i.

wow the 9260 is very slow. Reminds me of adaptec performance very much. Cant wait till Areca's new controllers whenever they are coming. Thanks for the testing!

Quote:

---

Originally Posted by Biker

Thanks for testing :)

As your results show the LSI card's 4k performance is hurt by it's SAS overhead (and possibly smaller cache) with higher latency meaning lower small file performance while the Areca falls behind at high sequential speeds due to it being PCI-e 1.0 and previous generation.

Informative results indeed :up:

---

You just stated the summary I should have put at the end of the first post of this thread - well said, thanks.

Quote:

---

Originally Posted by NapalmV5

1231 vs 9260
800mhz vs 800mhz <--- IOP 34x vs new PPC. Apples to Oranges. 3.0GHz P4 != 3.0GHz i7
sata2 vs sata2 <--- SATA2 vs SATA2 w/ SATA 3 support (won't change performance, but longevity)
hard vs soft <---- I won't necessarily dispute this, but I must have missed the thread that proved the 9xxx series is software-based for SATA
@ real apps theres no comparison <---- Proof?

---

Questioned points above.

I'm sure I missed something somewhere, but why are people suggesting this card does SATA in software and not hardware? I can't recall another SAS card before where this was the case, so I can't figure out why it would be now. It's just a different set of simple instructions to send to a drive, that's it really. If it turns out that it is done in software though I admit it's stupid... but in light of posts like the one I quoted where every other spec listed was questionable, I have to ask about this one.

Looks like updated firmware is still needed at least though, IMO. But eh, the same goes for every controller when it was new. Seeing as Intel is on the bandwagon as well I'm sure we'll see it sooner or later.

Edit: Can anyone with first hand experience let me know how hot that controller gets when idle and during load? What kind of cooling is required?

It gets pretty warm.

I'd say direct cooling would be mandatory unless you were running it in a server case with very high airflow.

I haven't got a temperature monitor on it directly but i'd guess it would go way over 60 centigrade under load if left passively cooled, much like the Areca.

It would be easy to strap or screw a 50 mm or 60mm low profile fan to it for direct cooling but this would make it dual slot.

Personally I have installed in my board's bottom slot and have a 120mm underneath blowing straight at it which keeps it very well cooled.

I would really like a temperature monitor implemented in the management software like with Adaptec cards, that would really be useful...

I haven't experienced any heat issues with my card at all...it runs very nice and cool, but i also run a torture rack so there is nothing BUT airflow. :)
I am glad to see some comprehensive testing between these two controllers, it is very informative!
I have recently tested the 1680-ix vs lsi-9260-8i personally, but not in near as scientific a manner! Kudos to steve-ro for thorough documentation.
Of course the 9260-8i did outstrip the 1680-ix by leaps and bounds, but the 1680 is SAS as well, so i think that is closer to comparing 'like' controllers.
It is clear that the 1231 does impressively well with small files, due to its lower latency, but i also question how the future implementation of TRIM will affect the performance of these controllers..i think that the 9260-8i is designed for this added instruction set, and has headroom to implement it. This is a controller developed with SSD in mind! The 9260-8i is a controller that is designed with future improvements in mind, improvement that have not come to fruition yet. I dont think that the 1231 has that much more room to improve, frankly, it is only going to have a very limited amount of future usage.
As to real world apps and the like i think that the people who are using these things care very little of real world usage. Who really uses these outside of us? We are the drag racers out here, and speed and being on the cutting edge is king(for some LOL)! They are both great controllers, i love areca, and when their 1800 series hits ill likely be one of the first in line to get it! BUT, this LSI is a damn fine piece of kit, and i refuse to believe that me getting 1.7 GB/S and NOT EVEN TOPPING THE CARD OUT stands for something impressive and special, something that thar 1231 cant even BEGIN to do!!! Imagine what this card will be capable of with NEXT GEN drives...it is a NEXT GEN controller...it is not optimized fully yet, because we arent even using it with what it is designed for yet!!
does anyone feel me on this? LOL
The last firmware revision gave me over 20 percent increase in small file transfer speeds, what will the next do?

Great responses guys! Very informative!
:)

just running some vantage guys...gonan get some justice here i guarantee it!
http://service.futuremark.com/compare?pcmv=222666
#9...oh and that is with the 512k stripe, which is not near as good as 1 MB!

So basicly your test shows that if you are only going with a few drives the 1231ml would be the better choice, and for more drives the 9260 due to higher maximum throughput?

The big drop in read performance in ATTO when the access pattern changes from 1*stripe size (64KB) to 2*stripe size (128KB) on both controllers is very odd.

The random I/O perf on the 9260 is very disappointing. Makes me think I should wait until Q1 next year and upgrade to the Areca 1800 series instead -- assuming they can solve the small-block latency issues, of course.

Quote:

---

Originally Posted by Anvil

SteveRo

Thanks for the comparisons.

Could you please list the parameters you used for the iometer tests?
I've got some Intel's I'd like to compare to your results using the LSI 9260-8i.

---

I apologize for not getting back to you sooner -
The settings I used in iometer are as you see them listed with everything else on default settings with each measurement run for 1 minute each.
One minute for 4MB 100% sequential 100% read, queue set to 64
One minute for 4MB 100% sequential 100% write, queue set to 64
One minute for 4KB 100% random 100% read, queue set to 64
One minute for 4KB 100% random 100% write, queue set to 64
...

Quote:

---

Originally Posted by AceNZ

The big drop in read performance in ATTO when the access pattern changes from 1*stripe size (64KB) to 2*stripe size (128KB) on both controllers is very odd.

Maybe due to running out of cache?

The random I/O perf on the 9260 is very disappointing. Makes me think I should wait until Q1 next year and upgrade to the Areca 1800 series instead -- assuming they can solve the small-block latency issues, of course.

---

Where did you see Areca 1800 not till next year?
That's dissapointing too.

The Areca 1800 is going to be a SAS card which means FAIL as far as SATA SSD latency & 4k performance is concerned.

I think that is signaling that it wil be awhile before we see any new sata controllers...

Quote:

---

Originally Posted by SteveRo

Where did you see Areca 1800 not till next year?
That's dissapointing too.

---

There was a video interview somewhere of a rep from Areca. He said it would be released Q1 2010, that they were using the Marvell chip, and that the price would be well over $1000 for the 8-drive version.

Quote:

---

Originally Posted by AceNZ

There was a video interview somewhere of a rep from Areca. He said it would be released Q1 2010, that they were using the Marvell chip, and that the price would be well over $1000 for the 8-drive version.

---

You mean the Saaya interview from a while back? I think that was a SAS controller he had in his hands. The man seemed to be very unsure of himself when asked about SSD's.

Quote:

---

Originally Posted by SteveRo

I apologize for not getting back to you sooner -
The settings I used in iometer are as you see them listed with everything else on default settings with each measurement run for 1 minute each.
One minute for 4MB 100% sequential 100% read, queue set to 64
One minute for 4MB 100% sequential 100% write, queue set to 64
One minute for 4KB 100% random 100% read, queue set to 64
One minute for 4KB 100% random 100% write, queue set to 64
...

---

No problem, I received the last drive yesterday so I haven't had time to do the tests yet.

What filelength did you use and how many workers?

For that kind of price... I don't know What they can offer that will be worth it over the new LSi card...

Quote:

---

Originally Posted by Anvil

No problem, I received the last drive yesterday so I haven't had time to do the tests yet.

What filelength did you use and how many workers?

---

file length was 8-10GB, workers was 1.

@ SteveRo

Nice comparison.
The Copy-Test and seq. r/w with LSI and Acard is disappointing me. :down:
LSI only a paper tiger?

I want to see more tests with real apps.
I think i can stay with Areca

Quote:

---

Originally Posted by F.E.A.R.

@ SteveRo

Nice comparison.
The Copy-Test and seq. r/w with LSI and Acard is disappointing me. :down:
LSI only a paper tiger?

I want to see more tests with real apps.
I think i can stay with Areca

---

Yes, the Areca is faster at random IO than the LSI no questions.
When using acards it makes a big difference.
That said, the LSI is faster at sequential reads - even with acards.

i would like to see some bootup runs with the 1231 on I/O meter, and the random file runs on I/o Meter as well. I guess that i remain unconvinced because these benchmarks are not reliable. How in the world, for instance, can the areca be getting 1635 mb/s seq read speed in the crystal disk benchmark??? that is physically impossible for that controller, and that is a widely accepted fact. The atto benchmark screens are totally unreliable as well with the controllers,l look at the 128.0 results..you expect me to believe that the 1231 is reading at 100 mb/s but writing at 850 at that block size? on other forums that i frequent there has been a agreement that the only real benchmarks that work with hardware raid is Everest and I/O meter. I would really really like to see runs with those if you have time stevero, i could provide you linkage to i/o meter if you need it.

Quote:

---

Originally Posted by Computurd

I guess that i remain unconvinced because these benchmarks are not reliable. How in the world, for instance, can the areca be getting 1635 mb/s seq read speed in the crystal disk benchmark??? that is physically impossible for that controller, and that is a widely accepted fact.

---

2GB cache. The bench is running in cache.

Just like some benchmarks posted for the LSI are unreliable, unless you use a larger test size. Do some runs at 50MB and then 1000MB on CDM and I suspect you will see the difference on a 512MB cache controller. You see people posting 1500MB/s using only 2 drives on some benchmarks because the bench is run in cache.

I am aware of that, and also that lots of benchmarks only measure the peak performance of the device, they do not give an accurate representation of the whole, that is why i only really trust everest and I/O meter personally. However, this also does not account for the atto readings, atto is a borked bench when it comes to hardware raid. AS SSD is also not accurate with hardware cards.

Quote:

---

Originally Posted by Computurd

I am aware of that, and also that lots of benchmarks only measure the peak performance of the device, they do not give an accurate representation of the whole, that is why i only really trust everest and I/O meter personally. However, this also does not account for the atto readings, atto is a borked bench when it comes to hardware raid. AS SSD is also not accurate with hardware cards.

---

I agree with you on iometer, it is the one I trust the most also.
I did a complete iometer compare of the LSI vs the Areca controller in the first posting of this thread, see the first table that I posted -

http://www.xtremesystems.org/forums/...d.php?t=234374

- iometer 2008.06.18-RC2 was used. The table (in the first post again) is the excel output from iometer.
All of the conclusions that I draw from the comparison are based on these iometer results.

Well that makes me feel dumber than usual, shoulda noticed it said i/o meter in the upper right corner there...:confused: for some reason i was thinking that was a compilation of the otyher benchmark scores...my fault!

im testing the Areca 1231ML
how are you getting those huge crystalmark readings with SSDs?!?

@ Mr dinos - good morning and congrats on your acquisition! :up:
If the CDM runs you are looking at are unbelievably high they are probably being run out of the areca controller cache.
When will we see your pcmv score - can't wait!

i only have stock cache 512MB on them
i hacked up one of my corsairs thats 2GB and killed the stick so lol have to test with 512MB
no wonder....you are running the bench on cache and im on actual drives hence the performance is :banana::banana::banana::banana:...same as adaptec

ok 2GB ram and i can see similar figures now..........but vantage is aweful :confused:

have u done any PCMV testing with SSDs?

LOL I got 24k with ONE ssd now :p:

awesome nizzen! which drive, and do you have it on your controller?