LSI 9265-8i

[alfaunits]

I talked with Ct on PM about this, so in case anyone else wonders (and wants to check the formula )
For Areca, since 520 sectors (512 byte sectors) are used at the beginning of the drive, that means 260KB is used from each drive.
If we wanted to align to say 100MB (102400 KB), we would need to create the partition at offset 102400-n*260, where n = number of effective SSDs in the array (i.e. don't include parity drives, if any).
Minimal offset is different for old and new SSDs, due to increased erase block sizes.

[Computurd]

excellent info, alfa! i really appreciate it

in related news....
They are sending the 1gb ddr3 daughterboard to me, my editor is the absolute best this should do wonders to help the Raid 5 and pcmv results

@Anvil--any word on your card? or are you off benching the heck outta it?

[felix_w]

CT, do you have any draft R-5 benches to show?

1GB Cache will take them to the skies as you said, but just for the rest to have an idea of what levels this card can reach...

[Anvil]

@Anvil--any word on your card? or are you off benching the heck outta it?

It's still waiting for customs clearance.

iWill, my vendor called me yesterday and informed me about the status, hopefully I'll be getting it by Friday as I'm going to work out of town next week.

So, all the benchmarks you've been running are done using only 512MB cache
That makes me wonder how the 1GB edition performs.

[Computurd]

yes i agree anvil. they are overnighting the module to me, it will be here in the AM. that will be very nice to test. we will see if it helps with the only "dark spot" which is the PCMV performance.

but really, who cares about PCMV lol. this thing is so ridiculously good! for regular usage (outside of benching) this thing is way better than the areca imo. i mean just from a 'using it' standpoint, it torches the areca.

for benchmarking of course the areca and its UBER cache own.

[Jose Teixeira]

yes i agree anvil. they are overnighting the module to me, it will be here in the AM. that will be very nice to test. we will see if it helps with the only "dark spot" which is the PCMV performance.

but really, who cares about PCMV lol. this thing is so ridiculously good! for regular usage (outside of benching) this thing is way better than the areca imo. i mean just from a 'using it' standpoint, it torches the areca.

for benchmarking of course the areca and its UBER cache own.

Good evening to all members,I'm new here.
Computurd, what do you mean thar LSI 9265 is way better than Areca? You refer to 1880ix12-4gb? How can it be better in real life aplications?
Thanks
Regards:)

[flamenko]

Wow....ow thats a lurker!!! First post since 09? Congrats!

[Computurd]

welcome jose! well, i guess youve been around a while

it is more about the off-the-blocks speed here than anything. The dual core of this ROC allows faster processing, and that is a big winner. The Areca 1880-IX has the LSI2108 ROC, which is the same as used in the LSI 9260. single core 800 mhz.
the lsi 9265 uses the next generation of that very chip, the LSI2208. dual core and paired with ddr3 memory, as the areca has ddr2.

so the question is this, is it faster? well the benchmarks (outside of PCMV) overwhelmingly say yes. but there are other metrics that are hard to quantify, and one is just the snappiness factor. how does it handle under big loads etc...

how about functionality, stability, and latency. all seem to be better to some degree. problem with the arecas is that without a battery if you BSOD you are going to lose your array a large percentage of the time. I think alot of it has to do with how much they leverage cache, and the different type of caching algorithms they use. Areca is definitely very cache dependent. they love to use the cache, and that is a good thing, and can be a bad thing.

LSI i have only lost one array, and that was on the 9260. and that was out of about 500 BSOD. and im not joking, hundreds of them! I overclock like a madman, so BSOD is a daily part of life for me when im doing bench runs. and you knwo what? ive never once had a BBU for an LSI. havent lost one on the 9265 either, and i have had a *few* bsod, and no BBU. would've definitely lost one by now if it were an areca.

I'm not trying to say bad things about areca. I LOVE arecas, but they have different usages and strengths/weaknesses than the LSI do. both controllers have their good points. The arecas allow online migrations and you can change raid types, stripe size and such, without even turning the thing off! amazing and easy that is a big strong point for the arecas. hell you can game while the card is changing the stripe size in the background. amazing!

in the terms of snappiness, with the new LSI it just seems faster. Sometimes when you are using things it comes down to that especially when doing heavy multitasking and such. I do admit that it will be rare occasions where you are going to push that hard, enough to notice a difference, for a normal person.

not to mention that the LSI is faster across the board in all specs. a big thing about areca is the cache and when you are gaming and reloading things repeatedly it is faster, because you are levering cache. but those are cases that you wont see very often. The LSI is going to be faster just off that FIRST load.

i havent really did a head-to-head comparison of them, just initial off-hand comparisons that i am making here of course. Not sure if i have the time to do a big comparison head to head right now, been very busy lately

and tbh, i have always felt that the 9260 with FastPath is faster than the 1880 in real life usage. the only reason i switched to the areca is that i bench alot, and that benches WAY better with a full complement of cache.

adn then there is scaling. the 1880 doesnt scale as well with large arrays. you leave alot of performance on the table, especially with random access. the 9265 has yet to be fully saturated in that aspect with the gear im using. 5 vertex gen 1 can saturate the 1880. 8 C300 cannot saturate the 9265.

depends upon usage i guess.
for me: benching=areca (so far, i havent tested LSI with the full 1gb ddr3 cache yet)
real life=9265 by a large margin.

wait until areca puts a 2208ROC on one of their cards with 6gb of DDR3...then we will see some major difference!

[alfaunits]

If you set no read-ahead and write-through, why would the cache size make ANY difference?
That said, if it does make a difference, aren't you suspicious as to why?

[Computurd]

If you set no read-ahead and write-through, why would the cache size make ANY difference?
That said, if it does make a difference, aren't you suspicious as to why?

if you take cache totally out of the equation, the 9265 wipes the floor with the 1880.

the LSI is worse with read ahead if you are using fastpath. no doubt about that. and write back as well if you are using raid 0, but if using R5 then it seems to be better with R5.


you cant really turn off the cache on the 1880. weird that. it uses it regardless it *seems*, and this is something other users have noticed. like the controller behaves like a dog with a 1gb stick, but with a 4gb stick it is a champ.

[Jose Teixeira]

if you take cache totally out of the equation, the 9265 wipes the floor with the 1880.

the LSI is worse with read ahead if you are using fastpath. no doubt about that. and write back as well if you are using raid 0, but if using R5 then it seems to be better with R5.


you cant really turn off the cache on the 1880. weird that. it uses it regardless it *seems*, and this is something other users have noticed. like the controller behaves like a dog with a 1gb stick, but with a 4gb stick it is a champ.

Hello everybody

Yes, I've never tested LSI 9260 or Areca 1880,but benchemarks of them show that Areca seems handle 4k ramdom read/write better than LSI 9260 and I think 9265 too,according to AS SSD and crystal disk mark.
The impressions of Areca 1880 is just that...always runs on cache,specially with 4gb module (with read ahead and write back cache). I think the areca uses it even runs aplications before the first time (read ahead) and put to cache (look for 150~200mb/s 4k random read). LSI 9260 can't do it... Well I'm a owner of a 3ware 9690SA and know this controller is much slower because firmware seems to be made for sequencial works, so in 4k random read/write QD1, looks like an LSI9260/61 and 3ware 9750, no more than 30mb/s. I would really like the best controller for 4k. Althought, on 9265 review's, I liked to see the 4k random QD32, almost 1gb/s...this represents the horsepower of dual core 2208 and, for me, this is the big advantage for 9265 vs 1880...am I right?
It seems hard what I choose...

[Jose Teixeira]

The great pseudo-performance is...4Gb cache...that's what a entusiast likes to think hehehe...
Computurd, I have a 3ware 9690 with BBU + 2x Corsair F120 (it works!) Raid 0.The BBU makes the big rule on this, in cache obviously but I noticed on installations of applications for the first time, really fast vs without BBU + write back cache.
Up to 512mb of data, the sequencial data is done at 1gb/s read and write, evens 512k too, according to CDM. 4k QD1...lol just 25mb/s and 40mb/s write...
Appart of this,I think more on Areca than the LSI. Speak Computurd...the guru of dark forces of controllers

[alfaunits]

the LSI is worse with read ahead if you are using fastpath. no doubt about that. and write back as well if you are using raid 0, but if using R5 then it seems to be better with R5.

I sure hope your tests were with RAID0? R5 changes the picture, and my opinion also (to a higher than previous positive)

you cant really turn off the cache on the 1880. weird that. it uses it regardless it *seems*, and this is something other users have noticed. like the controller behaves like a dog with a 1gb stick, but with a 4gb stick it is a champ.

We'll need to figure what "seems" is, since when I disable read-ahead or write-back for an array, I can't go over single-drive speeds with Areca, period.

[Jose Teixeira]

I sure hope your tests were with RAID0? R5 changes the picture, and my opinion also (to a higher than previous positive)


We'll need to figure what "seems" is, since when I disable read-ahead or write-back for an array, I can't go over single-drive speeds with Areca, period.

That's reason why areca is so fast,even if you disable read ahead or write back, she's gonne use it (not totally of course).
In my opinion, overall I think Areca did better job with LSI 2108 ROC and our firmware than LSI 9260, do you think the same?

[alfaunits]

That's reason why areca is so fast,even if you disable read ahead or write back, she's gonne use it (not totally of course).

No, it won't. Without read-ahead and write-back, you get ABYSMAL sequential performance.

In my opinion, overall I think Areca did better job with LSI 2108 ROC and our firmware than LSI 9260, do you think the same?

Our? You work for Areca? Better in what way exactly?

[Jose Teixeira]

No, it won't. Without read-ahead and write-back, you get ABYSMAL sequential performance.



Our? You work for Areca? Better in what way exactly?

My mistake sorry,lol,not our.
Well, in my card, if I disable the read ahead and write back, I'm don't have abysmal sequential performance. RH and WB helps to optimize sequencial,but this it's other case because firmware of 3ware 9600's series is very different than the Areca.

On topic:
Afaunits, I'm speaking just for what I see on Areca's benchmarks, that it seems handle the random performance better than LSI 9260...with read ahead and WB,even with only 1gb ram module...
But, ok, talk what you think
REgards

[alfaunits]

My mistake sorry,lol,not our.
Well, in my card, if I disable the read ahead and write back, I'm don't have abysmal sequential performance. RH and WB helps to optimize sequencial,but this it's other case because firmware of 3ware 9600's series is very different than the Areca.

Oh, I thought you meant Areca, not 3Ware. On Areca the RH and WB are required for good sequential performance.

Afaunits, I'm speaking just for what I see on Areca's benchmarks, that it seems handle the random performance better than LSI 9260...with read ahead and WB,even with only 1gb ram module...
But, ok, talk what you think
REgards

Ah, so of 3 things (low QD random, high QD random and sequential), Areca is good in one non-enterprise use, so it's awesome, yet it was made for enterprise use? (just joking, nothing personal; I gave up on Areca after too many lost arrays for no reason, seeing other people loosing arrays and it possibly frying 2 of my X25-E SSDs )

[Jose Teixeira]

Oh, I thought you meant Areca, not 3Ware. On Areca the RH and WB are required for good sequential performance.



Ah, so of 3 things (low QD random, high QD random and sequential), Areca is good in one non-enterprise use, so it's awesome, yet it was made for enterprise use? (just joking, nothing personal; I gave up on Areca after too many lost arrays for no reason, seeing other people loosing arrays and it possibly frying 2 of my X25-E SSDs )

Did you use a BBU before lost the arrays?

[alfaunits]

Did you use a BBU before lost the arrays?

I have a UPS, all array losses were after PROPER shutdown. The arrays just got lost on restart (whether reboot or shutdown and power-on next day).
Enough similar experiences are present on 2CPU forums if you want to confirm it, there's no doubt it looses arrays. Now, whether loss of a WD non-RE drive array should be scrutinized... maybe not, it's not s supported configuration. But loosing SSD arrays? Several times? Not good.

[Jose Teixeira]

I have a UPS, all array losses were after PROPER shutdown. The arrays just got lost on restart (whether reboot or shutdown and power-on next day).
Enough similar experiences are present on 2CPU forums if you want to confirm it, there's no doubt it looses arrays. Now, whether loss of a WD non-RE drive array should be scrutinized... maybe not, it's not s supported configuration. But loosing SSD arrays? Several times? Not good.

But you know that UPS is not a substitute of a BBU. It's completely different. BBU is an electronic module with a own CPU that retain data on the cache (with PC enable or not) and improves performance with basic read,read ahead and write back cache. The UPS don't do that, no way.

[Jose Teixeira]

Forget to tell other thing...the BBU also reorganized data when system is running or not. Course this is transparent for us. I can ensure you with an UPS without BBU and make a dirty shutdown or unplug one drive of the array and wait up to UPS turns off, certainly say goodbye to array. (I learned in the worst way). Made a experience with BBU,without UPS, the same steps, and not even show me a chekdisk on startup. Went to see alarms of a controller and cache was syncronized in a time that made a dirty shutdown or unplug the drive. 5 stars.

[Tiltevros]

Forget to tell other thing...the BBU also reorganized data when system is running or not. Course this is transparent for us. I can ensure you with an UPS without BBU and make a dirty shutdown or unplug one drive of the array and wait up to UPS turns off, certainly say goodbye to array. (I learned in the worst way). Made a experience with BBU,without UPS, the same steps, and not even show me a chekdisk on startup. Went to see alarms of a controller and cache was syncronized in a time that made a dirty shutdown or unplug the drive. 5 stars.

+1 on that

[Jose Teixeira]

Anyone has experience or just use the new Adaptec 6805 with a Zero Mantenaince Capcitor Module with 4Gb SLC?
Regards to all

[alfaunits]

But you know that UPS is not a substitute of a BBU. It's completely different. BBU is an electronic module with a own CPU that retain data on the cache (with PC enable or not) and improves performance with basic read,read ahead and write back cache. The UPS don't do that, no way.

And you do realize it does not matter since the shutdowns and reboots were clean, and the arrays just get lost?

Forget to tell other thing...the BBU also reorganized data when system is running or not. Course this is transparent for us. I can ensure you with an UPS without BBU and make a dirty shutdown or unplug one drive of the array and wait up to UPS turns off, certainly say goodbye to array. (I learned in the worst way). Made a experience with BBU,without UPS, the same steps, and not even show me a chekdisk on startup. Went to see alarms of a controller and cache was syncronized in a time that made a dirty shutdown or unplug the drive. 5 stars.

WOW awesome! WHAT cra* of controllers (other than Areca) looses an array if a drive is just unplugged? Or after a dirty shutdown??? Unless you do migration/expansion at the time, any controller that looses an array on a dirty shutdown is faulty!
They are SUPPOSED to survive s*** like that, it's their main purpose - REDUNDANCY and up time.
It must not even loose a RAID0 array, much less R1/R5/R6/etc.
It can have corrupted data on the array without a BBU, but NOT LOOSE AN ARRAY!!!!!!!!!!!!!!!!!!!

[Jose Teixeira]

And you do realize it does not matter since the shutdowns and reboots were clean, and the arrays just get lost?



WOW awesome! WHAT cra* of controllers (other than Areca) looses an array if a drive is just unplugged? Or after a dirty shutdown??? Unless you do migration/expansion at the time, any controller that looses an array on a dirty shutdown is faulty!
They are SUPPOSED to survive s*** like that, it's their main purpose - REDUNDANCY and up time.
It must not even loose a RAID0 array, much less R1/R5/R6/etc.
It can have corrupted data on the array without a BBU, but NOT LOOSE AN ARRAY!!!!!!!!!!!!!!!!!!!

I'll tell you. That experiencie was an a 3ware 9650. Although I shutdown the system when the controller was install an update of Windows, to see what happened.
On 3ware 9690, everything run fine, just did the checkdisk (with UPS, no BBU). Than, when I put the BBU and disable the UPS,did the same things and just run fine on a variaty of situations, the BBU was there to save everything.
Other important thing, is relationship between BBU and read ahead/write back.
Sometimes, on a specific cases,when the controller is reading ahead and put it data on own cache and than make the job as program/installation whatever is designed, the BBU retains some tracks of data and than (if the program/installation) has to be writen, it write on cache without touch on drive/drives/array. Just after to be writen to the cache,the data (random or sequencial) goes to the drive, more less in a sequencial mode. So, somethings are read and write without touch on array or drives. (I'm not saying everything).Well, if I'm questioned if the BBU improves performance? I say YES. So, with a controller like a Areca with 4gb, I believe that things *seems* runs on cache, I mean at the first time.

Regards man