On Tue, May 08, 2001 at 09:44:53AM -0600, Chris Jones wrote:
> I'm sure this is stuff I should know, but I've never been very good at
> SCSI performance tuning.
> 
> I've got an i386 with some SCSI disks, and I'm working on building a
> couple of RAID arrays.  The first array is up and working, and I don't
> much care how fast it is.  The second, though, needs to be speedy.
> 
> Partial dmesg output follows:
> 
> NetBSD 1.5.1_ALPHA (GAMERA) #0: Tue Jan 23 11:43:12 MST 2001
> cpu0: Intel Pentium II (Klamath) (686-class)
> siop1 at pci0 dev 11 function 0: Symbios Logic 53c875 (ultra-wide scsi)
> siop1: using on-board RAM
> siop1: interrupting at irq 10
> scsibus1 at siop1: 16 targets, 8 luns per target
> siop0 at pci0 dev 14 function 0: Symbios Logic 53c875j (ultra-wide scsi)
> siop0: using on-board RAM
> siop0: interrupting at irq 9
> scsibus0 at siop0: 16 targets, 8 luns per target
> scsibus1: waiting 2 seconds for devices to settle...
> siop1: target 0 using 8bit transfers
> siop1: target 0 now synchronous at 20.0Mhz, offset 16
> siop1: target 0 using tagged queuing
> sd2 at scsibus1 target 0 lun 0: <IBM, DMVS36D, 0210> SCSI3 0/direct fixed
> siop1: target 0 using 16bit transfers
> siop1: target 0 now synchronous at 20.0Mhz, offset 16
> sd2: 35003 MB, 11739 cyl, 20 head, 305 sec, 512 bytes/sect x 71687340 sectors
> siop1: target 1 using 8bit transfers
> siop1: target 1 now synchronous at 20.0Mhz, offset 16
> siop1: target 1 using tagged queuing
> sd3 at scsibus1 target 1 lun 0: <IBM, DMVS36D, 0210> SCSI3 0/direct fixed
> siop1: target 1 using 16bit transfers
> siop1: target 1 now synchronous at 20.0Mhz, offset 16
> sd3: 35003 MB, 11739 cyl, 20 head, 305 sec, 512 bytes/sect x 71687340 sectors
> st0 at scsibus1 target 5 lun 0: <HP, C1557A, U812> SCSI2 1/sequential removable
> st0: siop1: target 5 now synchronous at 10.0Mhz, offset 16
> density code 37, variable blocks, write-enabled
> ch0 at scsibus1 target 5 lun 1: <HP, C1557A, U812> SCSI2 8/changer removable
> ch0: 6 slots, 1 drive, 0 pickers, 0 portals
> scsibus0: waiting 2 seconds for devices to settle...
> siop0: target 0 using tagged queuing
> sd0 at scsibus0 target 0 lun 0: <SEAGATE, ST34572W, 0784> SCSI2 0/direct fixed
> siop0: target 0 using 16bit transfers
> siop0: target 0 now synchronous at 20.0Mhz, offset 15
> sd0: 4340 MB, 6300 cyl, 8 head, 176 sec, 512 bytes/sect x 8888924 sectors
> siop0: target 1 using tagged queuing
> sd1 at scsibus0 target 1 lun 0: <SEAGATE, ST32272W, 0784> SCSI2 0/direct fixed
> siop0: target 1 using 16bit transfers
> siop0: target 1 now synchronous at 20.0Mhz, offset 15
> sd1: 2157 MB, 6300 cyl, 4 head, 175 sec, 512 bytes/sect x 4419464 sectors
> 
> In a nutshell, the problem is this:  sd2 and sd3 are getting a
> throughput rate of about 1MB/s.  I have another disk in the office,
> though, which can get about 9MB/s.  (This is measured with "dd
> if=/dev/zero bs=1024k count=100 of=/dev/sd?d".)  The other (fast) disk
> is a year-old Seagate Barracuda, if that helps any.  I'm using the
> same enclosures and cables, so I just don't see what would make the
> difference here.  To make matters more confusing, I have another,
> identical Barracuda which runs at 1MB/s.
> 
> Any thoughts on what's going on here?

Do the scsi controllers share interrupts with other devices ?
While testing disks speed, could you run a 'systat vm' and look at irq rates ?
Do you get decent read speed out of these disks ?

--
Manuel Bouyer <bouyer@antioche.eu.org>
--