Question about the LSI SAS 9207-8i

[Tybio]

I've just upgraded to 2 of these from 2xAOC-SAS2LP-MV8 as I'm planning on upgrading to a Threadripper later this summer and setting up a windows VM.  While the speeds are OUTSTANDING, I'm struggling with a new issue where the disks become unavailable when another spins up.

 

If I'm streaming from disk4 for example, when something spins up disk2 the video pauses and loses sync as it seems access to the data freezes until disk2 comes fully online.  I suppose I could solve this with increasing the buffering, however it strikes me that I shouldn't expect this behavior on a PCIE 3.0 HBA....Unless I'm missing something?

 

 

[Tybio]

It would be really nice if anyone could chime in here.  I'm not sure if this is a known limitation of the LSI cards, or if there is something else going on.  I'm actually considering returning them and going back to the SAS2LPs and praying things don't blow up when I setup my windows VM.

[uldise]

just for curious: are you on latest firmware on 9207?

[Tybio]

As far as I can tell, they are both current:

 

root@Storage:~# dmesg | grep mpt2sas | grep FWV
[   13.714606] mpt2sas_cm0: LSISAS2308: FWVersion(20.00.07.00), ChipRevision(0x05), BiosVersion(07.39.02.00)
[   14.234291] mpt2sas_cm1: LSISAS2308: FWVersion(20.00.07.00), ChipRevision(0x05), BiosVersion(07.39.02.00)
root@Storage:~#

 

[Tybio]

No luck at all with this, going to swap back to my AOC-SAS2LP-MV8 to see if that suffers the same issue.  All I can think of is that something on the MB isn't happy, going backwards is the only way I can test it out...so we will see.

[Tybio]

And it works perfectly fine on the AOC-SAS2LP-MV8, I can spin up/down disks and even spin up/down all disks and not a dropped frame.  I guess I'm just a SuperMicro person from here on out.  I'd consider selling on those LSI cards, but I don't want to risk passing on my problems to anyone else...guess I just eat the cost.

[electron286]

  I am not sure, BUT I think there may be something with the combination of the specific driver in use with the drive controller, PCI-e chips on the mother board and drivers, Bios implementations (mother board and drive controller), and even interrupt processing going on here.  Long list I know.

 

  I have done quite a bit of testing with various setups over the years, and have found both the SuperMicro AOC-SAS2LP-MV8, AND the LSI SAS 9207-8i/e cards to be very capable and stable.  However that has not been the case for many users.  When starting initial configurations for testing, I do (usually) first check for which slots on the mother board provide the highest bandwidth, and least amount of chip to chip translation of data.  I have seen on some motherboards that one slot will work better than another, and that depending on the controller, it may be more usable in a different slot than another control design.

 

  The interesting thing is that many people have similar symptoms to what you were seeing when running the SuperMicro AOC-SAS2LP-MV8, so they have switched to the LSI SAS 9207-8i/e cards with no issues at all!

 

  It seems that the general feeling on most linux based system forums is that the LSI SAS 9207-8i/e cards are the more stable and usable ones over the SuperMicro AOC-SAS2LP-MV8 cards.  It does also seem that the LSI SAS 9207-8i/e cards, especially the 8e, are faster and more expandable and with lower performance impact, when using SAS port multiplexers/expanders.

 

  Sadly, with all the "compatible" and interchangeable hardware out there, that we can all swap around and mix and match, there will always be a potential for incompatibility.  For example, I have had more erratic performance issues when trying AMD based systems than Intel based systems over the years (since the K6-2 was released, prior to that the AMD processors had no negatives compared with Intel processors in my tests).  AMD systems are very usable for most applications, just typically not as consistent in performance, a carry over from the "enhancements" made in the K6-2 processors to allow for faster processes to have streamlined paths for an overall performance boost, at the cost of repeatable and predictable output speeds.