LSI 9211-8i versus LSI 9260-8i versus Areca 1231ML-2G

[trans am]

Computurd explained to me that when since we are setting up with adaptive read ahead your system needs to learn how you are using the computer and will adapt to your usage over time. performance will increase.

[Computurd]

i was not referring to the adaptive read ahead...the 9260 has an advanced algorithm that controls how it uses its cache...it basically tailors it to how you use yo9urr system, so there is a "learning cycle" that it uses, as a matter of fact there is even options for that in the MSM. also, if youhaveread any reviews about the 9260 they always mention that the algorithms give them a pain in the ass when they try to bench the card.

[trans am]

hit me on google > I was way off!

[minhgi]

These benchmarking does not look right..
my setting so far:

Raid 0 4 Supertalent Ultra 32 GB
Adaptive Read Ahead
Write back
64kb block
IO Cache
Cache enable

Need help if possible...

Notice having IO cached enable, decrease my boot time by 5 second. With Direct IO, it still give me the same bar graph (scraping at 64k) and beyond. Should I increase the block to 512KB?

[Computurd]

going to be testing the 9260-8i new FW very shortly...had some RLS to take care of...

[idolclub]

The new firmware of the 9260-8i has a great improvement!

[Computurd]

i dont think you have seen anything yet from these controllers (lsi). the game is about to change.
it will become very one-sided

[SkOrPn]

i dont think you have seen anything yet from these controllers (lsi). the game is about to change.
it will become very one-sided

Comp, what do you mean by this statement? Is there a new raid chip out or is a new one in development? Or do you mean these controllers have massive amounts of improvement through new firmware releases?

[Tiltevros]

i came across with this PDFs from areca that as an IT makes me wonder for some things. This controller performs only with cache?? isnt a controller that Enterprises can work without All in cache?? there is no performance in Critical Data??? because in raid50 r in raid60 some companies not only the LSI r way way faster!!

Is this controller only for benchmark enthousiast?? Is this controller only for RAM-Cache Benchmarking??
And if is that.. Why should i Pay 600euros for 4 gb of cache???

[SteveRo]

I think the idea is you use battery backup and your ok?
Also - i can confirm - the 1231 will not boot w/o the memory in the slot - I have tried.

[AceNZ]

Can anyone summarize the differences in the performance and features of the 9211-8i, 9260-8i and the 1231ML-2G, when used with SSDs such as the X25-M?

[SteveRo]

AceNZ - all three should work very well with the x25-m.
General assessment - I have seen some huge iometer runs with the 9211 and 8x x25-m - huge throughput and with small file reads the highest anywhere - maybe with the exception of the ioxtreme.
In applications where cache can be used - both the 9260 and the 1231 will do well, usually the 1231 will do a little better as at has a larger cache (512MB vs 2 or 4GB)
The 1231 is now old and has a ceiling of 800-900MB/s - so 4 x25-m would max out the 1231 is seq reads.
We have seen at least one vendor advertise the new Areca 1880 to be available in late May, many folks are waiting to see what that looks like.

[AceNZ]

Rumor is that the 1880 will be well over $1000, so it would have to be pretty amazing to be worth considering. Also, although I've heard good things about Areca's UI, it seems like their firmware updates have been very infrequent compared to LSI.

How does the 9260 perform when cache is disabled? Can it approach the same level of IOPS as the 9211? The 9211's throughput is attractive on the surface, but its limited configurability strikes me as a potentially serious handicap in real-world scenarios.

[GullLars]

I have suggested in another thread that we should collectively query LSI wether a firmware update to enable configuring stripe size (and other things people may want wich is avalible on 9260) is possible. If it is, i would think LSI would get better rep for releasing such a firmware, even if it's only as a beta for us enthusiasts.

[AceNZ]

When LSI says that the 9211 has a fixed 64K "stripe" size, they must really mean strip size, not stripe size, right?

If so, it seems like that might be part of the reason why random IOPS are good, while I/O for "medium-sized" files isn't as good as the 9260 -- since random I/O would tend to use all drives at once (given a large enough QD), while file I/O would tend to focus I/O more heavily on a smaller number of drives (64K strip x 8 drives = 512K stripe).

I wonder how the performance of the 9211 compares to a 64K strip size on the 9260 (with and without caching enabled)?

[Computurd]

man oh man we need to revisit this with fastpath on the LSI

[mbreslin]

\I wonder how the performance of the 9211 compares to a 64K strip size on the 9260 (with and without caching enabled)?

I can possibly dig up some 9211 benches and see if I have a direct comparison with 9260 at 64k. Give me a couple days. (8x 256gb crucial c300)

[Computurd]

mbreslin for the love of christ you need fastpath! it is just amazing, the ultra low latencies! i have never seen 4k scale so beautifully, even ich10r is put to shame. i'd love to do a pcmv run with that and your uber array

[mbreslin]

Your numbers look good sir. I'm sort of interested myself, definitely not 160$ interested.

Banana them.

[GullLars]

Hmm... You have a $3000+ array, and you are not interrested in $160 extra for possibly a 50-100% max performance boost? Compared to the total cost of your rigg it's pocket change.
Possibly computurd could overnight you his key so you could try it? (if that is possible)

BTW, lower latencies means a speed boost even at low queue depths (and singletasking), so you don't have to be in the 100Ks of IOPS to see a benefit. It boosts the "cpu accelerator effect" of your SSDs.

[Anvil]

mbreslin,

Get the Key

It looks like its worth the $ unless you've converted to the ICH

[mbreslin]

I just think it's a crappy way to do business.

I see crucial has put the firmware update back on their site so I guess I'll order the silly key and do the firmware flashing+new array all at once.

@GullLars: You've hurt my ego when pricing my array, can you please tell me where I can purchase 8 256gb c300s and a 9260 for anywhere in the ballpark of 3000$? If this was even possible I'd buy another 8 just to have 16 drives ready for one of the 1880s.

edit: Best place to order from in the states? LSI directly or?

[Computurd]

lol yoiua re correct mbreslin, i was thinking the same thing when i read that, WAY more LOL.
scsi4me has them and they are awesome so thats where i would recommend getting it

[SkOrPn]

I agree with mbreslin 100%, not only is it crappy business practice, they should be sued for selling a product that is intentionally capped and then telling you if you want the full benefits of what the product can achieve you must pay extra. Sell the damn product with the key already included and charge more for the product in the first place, jeez. This business tactic is pure greed...

[mbreslin]

Ordered. I feel like such a sucker!

Edit: I should add I have never thought it was something they planned from the beginning only that they figured out how to get more out of the card in software after the product shipped. The price point seems high considering most other companies release their improvements via free firmware/drivers/software.