Optimum dev_openings value
Matthew Jacob
mj at feral.com
Thu Mar 18 17:06:01 UTC 2010
Openings changes reflect when and how a device sends back a QUEUE FULL
message, which causes FreeBSD to adjust the openings.
That is disk, access pattern, and timing dependent.
There are a number of theories about whether the 'openings' value should
be bumped back up after having been throttled down. Nobody can agree on
these theories, so the openings stay where they are. You might try
manually adjusting them up on a periodic basis.
> Hi,
>
> I have three machines running 8.0-RELEASE, each with two mirrored Seagate
> SCSI disks using ZFS. In making sure these machines are performing as
> best as possible, I'm trying to find what the best practices are with
> regards to dev_openings values, and what determines the default
> dev_openings value.
>
> Two machines have identical ahc(4) Adaptec AIC7899 Ultra160 SCSI adapters,
> but display different defaults, which I have not had luck finding an
> explanation:
>
> h04# camcontrol tags da0 -v
> (pass0:ahc1:0:0:0): dev_openings 33
> (pass0:ahc1:0:0:0): dev_active 0
> (pass0:ahc1:0:0:0): devq_openings 33
> (pass0:ahc1:0:0:0): devq_queued 0
> (pass0:ahc1:0:0:0): held 0
> (pass0:ahc1:0:0:0): mintags 2
> (pass0:ahc1:0:0:0): maxtags 255
>
> h21# camcontrol tags da0 -v
> (pass0:ahc0:0:0:0): dev_openings 64
> (pass0:ahc0:0:0:0): dev_active 0
> (pass0:ahc0:0:0:0): devq_openings 64
> (pass0:ahc0:0:0:0): devq_queued 0
> (pass0:ahc0:0:0:0): held 0
> (pass0:ahc0:0:0:0): mintags 2
> (pass0:ahc0:0:0:0): maxtags 255
>
> h04# camcontrol devlist
> <SEAGATE ST318406LC 0109> at scbus1 target 0 lun 0 (pass0,da0)
> <SEAGATE ST318406LC 010A> at scbus1 target 1 lun 0 (pass1,da1)
>
> h21# camcontrol devlist
> <SEAGATE ST3146855LC 0003> at scbus0 target 0 lun 0 (pass0,da0)
> <SEAGATE ST3146855LC 0003> at scbus0 target 1 lun 0 (pass1,da1)
>
> Is there an obvious explanation for the difference here?
>
> The third machine uses sym(4), which I expect would cause a different
> default dev_openings (currently defaulted to 44).
>
> Is there a general rule to adjusting this value?
>
>
More information about the freebsd-scsi
mailing list