Date: Fri, 28 Feb 2014 12:02:31 +0100 From: Borja Marcos <borjam@sarenet.es> To: "Steven Hartland" <killing@multiplay.co.uk> Cc: freebsd-scsi@freebsd.org Subject: Re: ServeRAID M5210e passthroughand syspd corruption Message-ID: <886C951B-AB52-44C6-A9CB-955222C601DA@sarenet.es> In-Reply-To: <E89DEC8D40554EBAA2099036DA885B2D@multiplay.co.uk> References: <E8F5E131-F51D-448A-9035-646FEFF686F3@sarenet.es> <0B1A8780A28942B78E8FA4E3CE5EAE3E@multiplay.co.uk> <C8396C8A-0500-4626-886B-5C83A733DA82@sarenet.es> <E89DEC8D40554EBAA2099036DA885B2D@multiplay.co.uk>
next in thread | previous in thread | raw e-mail | index | archive | help
On Feb 28, 2014, at 11:45 AM, Steven Hartland wrote: >> I know, unfortunately this brain dead (unless proven otherwise) IBM >> machine doesn't seem to allow me to use a proper HBA. >=20 > Not PCI slots at all then? No, it was purchased in a funny configuration. Well, would be great if = it worked. The PCIe raiser card which would offer three slots is replaced by a = contraption with two disks and a LSI HBA (that's one of the "mps" ones but just with one = connector). The front of the machine has 24 slots for 2.5" disks. IBM makes them so = that you can use the two "back" disks for the OS, with a simple mirroring support, and = all of the front mounted disks for storage.=20 The concept is nice, *but* if only the "Invader" (that's indeed a proper = name!) could behave like a good HBA... >=20 >> That said, since I asked for assistance in 2007 and Scott Long = explained >> how to turn an aac card into a passthrough device I have machines = running >> like a charm. I am sure I can trust ZFS to detect any problems due to = that >> setup. >=20 > Some older LSI controllers (both these are LSI underneath) had both a = RAID > and HBA firmware however new cards don't have this option. I've done it with aac cards with the same results. Seems the bad apple = is this "Invader" thingy. >> When you said you're seeing mfisyspd corruption what specifically are >> you seeing? >=20 >> I described some of the problems several days ago, in short, either = using >> ZFS or UFS on one of those "mfisyspd" devices or a "da" created by = allowing >> passthrough to work and running a benchmark, I get data corruption. = No data >> corruption happens if I create a "raid 0" device on one disk.=20 >=20 > Specifics on this would be interesting, as there used be corruption = issues > with > 2TB disks on MFI but we fixed that quite some time ago. All of the disks are smaller than 2 TB. The exact models involved are: Samsung SSD 840 BB0Q (1 TB) OCZ-VERTEX4 1.5 (500 MB) and for the hard disk, a Dell branded Seagate (ST9146803SS) >> I've tried with Samsung and OCZ SSDs and, just wondering if it might = be a >> SSD specific problem of some sort, a Seagate disk. No luck. >=20 > We have a few machines in production here using MFI in passthrough via > mfisyspd0 on ZFS and haven't seen any issues. In our case its a = ThunderBolt > not a Invader generation card so if could be there's a problem with > Invader chipset support. The problem with mfisyspd (anyway, the Invader is corrupting using da = and mfisyspd) is that, as far as I know, it won't support TRIM nor quircks (such as 4K blocks) for = SSDs. Anyway, as I said, neither the syspd nor the "brute force" approach are = working) >> With so many manufacturers (even Sun!) tending including these = controllers >> on their motherboards, and even refusing to sell alternative options = (who >> wants a "software RAID" when you can get "a hardware one", they say) = I >> think that it would be good to make sure that passthrough mode is = supported >> in the best possible way. >=20 > We just ensure we go with HBA card if required. Yes, I think the solution will be to kick "turnkey, Windows oriented" = manufacturers such as Dell and IBM and specify the servers to the component level. Again, from my = experience, it can be quite a pain.=20 >> Maybe we could lobby the LSI Overlords, the manufacturers, both, or = just >> try to make the best possible use of the most commonly found = hardware. >=20 >=20 > Can't say we've had an issue, Dell, Supermicro all have options for = HBA > you just need to know the model numbers, which can be quite confusing. Of course. We've had some problems with Dell, though. I remember telling = a Dell sales rep to just shut up and serve the card I have ordered, period. = Sigh. > One chassis to avoid is the Dell C6220 series as it fails on 6Gbps = speeds > due to bad wiring. Thanks, useful information!! Borja.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?886C951B-AB52-44C6-A9CB-955222C601DA>