LSI 9260-8i - poor random read/write performance with SSDs?

lunadesign

Member
Aug 7, 2013
120
9
18
When you leave for a month the rats come in and take over :)
You're too funny!

RAID0, a multi Ghz CPU on a Mobo can do this faster than a few 100 Mhz ROC
As RAID0 requires no real processing, just a matter of storing data in a stripe across drives

HW controllers like the SAS9260 are for enterprise use ie SAS drives which are slower but more reliable.
Mobo controller can't do SAS
Mobo controller can't do more than 2x SATA3 (Z87 can do 6) off the controller for RAID
Mobo controller has no HW caching
Mobo controller has no error checking correcting
Mobo controller can't do RAID6, or spanned RAID5 (ie 50/60)

Try and take a Mobo created RAID set to another computer, LSI RAID will work on all LSI RAID cards.
Trust me....I'm totally on board (excuse the pun) with using a HW RAID controller. I'm just a bit surprised that a dual core 800MHz ROC would be a bit of a bottleneck when it should be doing very little other than passing the reads and writes through. Otherwise, I'd think we'd have to worry about the speed/power of the SSD controller. My guess is the LSI firmware has a bug and is doing more processing than it should be in this case.
 

mobilenvidia

Moderator
Sep 25, 2011
1,773
63
48
New Zealand
It's not the ROC, it can crunch RAID6 with ease

Simple tasks will be quicker by lower latency onboard controller
Lets see what happens with you attach 8x SSD's to the Mobo and compare with 8x on the LSI, oh wait you can't
HW RAID controller are for 8x drives not 2x, Intel will always be faster with 2x
 

lunadesign

Member
Aug 7, 2013
120
9
18
It's not the ROC, it can crunch RAID6 with ease

Simple tasks will be quicker by lower latency onboard controller
Lets see what happens with you attach 8x SSD's to the Mobo and compare with 8x on the LSI, oh wait you can't
HW RAID controller are for 8x drives not 2x, Intel will always be faster with 2x
So basically what you're saying is that what I'm seeing is typical because of the latency difference?

I assume this means I'll see the same with other vendors (ie, Adaptec, Areca, et al)?
 

swagler

Member
Feb 15, 2018
30
1
8
Our server fleet is all samsung SSD 840 PRO, they all have physical drive write cache enabled. They are solid.
mrkrad, could you please tell me if your 840 PRO's are receiving auto UNMAP commands. I've found a utility to manually UNMAP my 840 PRO 512, regain drive space, but haven't found a way to configure 9211.4i IT to auto UNMAP.
 

mrkrad

Well-Known Member
Oct 13, 2012
1,244
52
48
No I've just decommissioned all my 840 pro's/850 pro's.. They were running esxi 6.0 which does not support auto-unmap/shrink especially with a 9260 series raid controller in raid-mode.

None the less. I can tell you that the dl360 g9 with sm863a ssd's are about 8 times faster with the p440ar raid controller than the best OP 30% samsung 840/850's on g6/g7 servers. The 9260 is butt-slow, and so is pci-e 2.0 x8!

same size and # of ssd's - just newer cpu/backplane/raid controller = 8 times faster, in raid-5 no less! with 0% OP from the sm863's solid as heck!
 

swagler

Member
Feb 15, 2018
30
1
8
mrkrad, thank you for reply. I have recently installed a LSI 9211-4i in a 2005 Asus A8N-SLI Premium to gain 6mb/s transfer speeds for my 840 PRO 512. I found the hard way SAS2004 IT Mode doesn't support TRIM. Through further research I found that SAS utilizes UNMAP, TRIM being SATA counterpart. I've read mass blogs, it appears that a filter driver configured to send UNMAP commands is required, running in conjunction with Storport driver. This sounds to me like we're responsible for writing such a driver, or have done by a Server Guru, for I can't find any preconfigured setup. To Avoid removing SAS Controller, I found a utility called PerfectStorage 3.0. Looks like a disk defrager, but very fast, that manually unmaps all unused blocks. It works slick, but only for mounted drives, drive needs connected to motherboard to do full Magician TRIM on unformatted, boot... areas. I'll continue researching, I'm a C Programmer, but don't risk my data fooling around developing such a driver.
 

Evan

Well-Known Member
Jan 6, 2016
3,101
517
113
None the less. I can tell you that the dl360 g9 with sm863a ssd's are about 8 times faster with the p440ar raid controller than the best OP 30% samsung 840/850's on g6/g7 servers. The 9260 is butt-slow, and so is pci-e 2.0 x8!
Totally as expected and proper enterprise SSD with a decent raid controller in front. They may be SATA and not SAS but those SSD’s are decent.
 

mrkrad

Well-Known Member
Oct 13, 2012
1,244
52
48
mrkrad, thank you for reply. I have recently installed a LSI 9211-4i in a 2005 Asus A8N-SLI Premium to gain 6mb/s transfer speeds for my 840 PRO 512. I found the hard way SAS2004 IT Mode doesn't support TRIM..
the problem with trim is synchronous or asynchronous trim, and ssd's which corrupt themselves (840?!) when using asynchronous trim. Which is what caused a bunch of problems with linux when they initially assumed every ssd supported it! Synchronous trim is safe, but slows down massively the job queue when trying to batch many I/o's then stop for a synchronous trim, then resume. Crashed a bunch of older san's as well with esxi (5.5). running a batch job synchronously may be okay for some folks to slow down their in-process I/o's for 20-30 seconds but for others they need to have it run asynchronously to not stop in place I/o's in flight.

Which is probably why most folks haven't attempted to write their own raid-controller trim driver!