Harald Schmalzbauer
2012-Nov-23 13:39 UTC
cam SCSI negotiation issues (mpt in that case), only 3.300MB/s transfers
Hello, I have a LSI20320IE, wich is a LSI1030 (U320-SCSI, supported by MPT) behind an PCI-X-PCIe bridge. I'm using that for two tape drives: LTO-3 and DAT72. My (solved, see below) first problem was, that passthrough of that HBA with ESXi5.0 didn't work. As soon as mpt was kldloaded, there was a heavy interrupt storm detected (>100k irq/s) and throttling occured. No matter what IRQ it was assigend and no matter if it was shared or not. The dmesg output hasn't shown any info about MSI usage, so I told ESXi to use legacy interrupts in host mode instead of msi. Didn't help anything. After I verified that this passthrough configuration works well on other OS, I tried out some things and found out that hint.mpt.0.msi_enable="1" solves the interrupt storm problem, although dmesg output still is exactly the same: mpt0: <LSILogic 1030 Ultra4 Adapter> port 0x4000-0x40ff mem 0xd6440000-0xd645ffff,0xd6420000-0xd643ffff irq 18 at device 0.0 on pci3 mpt0: MPI Version=1.2.15.0 Btw, I'm still curious what the sysctl "hw.mpt.0.role" means? With my LSI1030 it's "3", an onther 1068, it's "1" Now after I got passthrough working, I first tried to saturate my LTO-3. With the help of mbuffer (zfs send -R DATASET/filesystem at totape | mbuffer -s 1m -t -m 500M -P 80 -f -o /dev/nsa1) I saw 60-120MB/s transfer rates, avaeraging at 100MB/s at moderate interrupt load (~1k). So for so fine. But there are problems with the DAT72 drive. The SCSI negotiation seems to fail: sa0 at mpt0 bus 0 scbus3 target 0 lun 0 sa0: <SEAGATE DAT DAT72-052 A16K> Removable Sequential Access SCSI-3 device sa0: 3.300MB/s transfers Now I read here http://lists.freebsd.org/pipermail/freebsd-current/2006-May/063603.html that this may be just cosmetic, but in my case I can't get more than ~3MB/s to/from the tape. Once again I changed the controller to another guest OS and got 5,5MB/s avarage write transfer rate with ntbakup (wich is sensible for the gained compression ratio). lsiutil.exe shows negotiation parameters with 80MB/s (40MT/s 16bit), which correponds to the drive's interface. Back in FreeBSD, I had no luck using "camcontrol negotiate sa0 -R 40" to alter the negotiation parameters. It always shows 3,300MB/s and seems to reflect reality. I'd higly appreciate any help, since I have no idea how this could be fixed. Slot-constraints limit my choice of SCSI-HBAs to PCIe ones... Thnaks, -Harry -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 196 bytes Desc: OpenPGP digital signature URL: <http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20121123/78fe7b35/attachment.sig>
Harald Schmalzbauer
2012-Dec-05 20:04 UTC
cam SCSI negotiation issues (mpt in that case), only 3.300MB/s transfers
schrieb Harald Schmalzbauer am 23.11.2012 14:39 (localtime):> ... > found out that hint.mpt.0.msi_enable="1" > solves the interrupt storm problem, although dmesg output still is > exactly the same: > mpt0: <LSILogic 1030 Ultra4 Adapter> port 0x4000-0x40ff mem > 0xd6440000-0xd645ffff,0xd6420000-0xd643ffff irq 18 at device 0.0 on pci3 > mpt0: MPI Version=1.2.15.0 > Btw, I'm still curious what the sysctl "hw.mpt.0.role" means? With my > LSI1030 it's "3", an onther 1068, it's "1"Any hint for me what "hw.mpt.0.role" means?> ...I had no luck using "camcontrol negotiate sa0 -R 40" to > alter the negotiation parameters. > It always shows 3,300MB/s and seems to reflect realityI'd still need help here if possible. Is there any point to start for non-developers? Any SCSI command "secrets" I can try? If anybody is interested in solving, but lacks hardware, I'd donate one LSI20320IE. Thanks, -Harry (not subscribed to freebsd-scsi@) -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 196 bytes Desc: OpenPGP digital signature URL: <http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20121205/efcafad9/attachment.sig>