So it sounds like you're saying that performance was degraded when you
had all the drives and the picker sharing the same channel but when you
put the picker on a separate card/channel the performance increased?
We discovered that we were in fact running both libraries on the same
SCSI bus! I think we must have been under the impression that we were
running each library on its own separate bus because each was on its own
card/PCI slot, but in fact having two separate PCI slots does not infer
two separate buses. We had never considered this before. We moved the
storage node to a different server where each library now has its own
dedicated bus. However, each library is still running on one and only
one card. So in the case of the ATL library, we have two SDLT drives and
the picker sharing channel A on one dual-channel Adaptec card on one
bus, and the Storagetek (it has 4 LTO drives; two drives on channel A
and two drives and picker on channel B) uses the second dual channel
card on the other bus. So at least we have the two libraries now on
separate buses. We're going to see if this makes a difference. We don't
have a third card, but we will be buying one. Once we have the third
card, we can then move the Storagetek's picker to its own card on a
third bus. We're not using channel B, though, on the card that the ATL
is using, so we could move the ATL picker to that channel and have the
two SDLT drives continue using channel A. Guess that couldn't hurt, but
we'll need to re-cable the libraries for any these options.
Do you think we need to have the pickers on their own channel or
actually have them on their own bus apart from the buses that the drives
are using?
I appreciate your feedback. Will keep the list updated on our findings.
I'm hoping that having the libraries on separate buses will help, but we
can go finer grade by moving the picker to yet another actual bus. I
guess we could have the ATL library's picker on a fourth bus also, and
maybe even split up the drives on the Storagetek so no more than two
share the same bus. That would require a total of 5 buses and 5 cards,
but not sure that that will be necessary. It seems like performance
could only be better, but maybe we're not pushing the drives enough to
warrant that?
Will keep everyone informed as the events unfold.
Thanks again.
George
Olaf Zaplinski wrote:
>
> Hi George,
>
> I can not tell much but my own Storagetek experience, here we have an L700e.
>
> When the jukebox is connected to the same HBA (Adaptec, 160 MB/s capable) as
> our two tapes on the second SCSI chain, then the 1st tape is logged on as 40
> MB/s + 8 bit, the second one as 80 MB/s + 16 bit. The latter one is correct.
> When I connect only those two drives to the Adaptec, everything is fine.
>
> This is why we drive the JB device with its own host adapter (Adaptec) and
> the two SCSI tape drive chains on others (LSI 53c1030). The L700's firmware
> is brand new, we had it serviced last week.
>
> So I think it is better to spend some money for a host adapter that is only
> connected to the jukebox/the picker. Since there is no heavy SCSI traffic, I
> guess that any 32 bit PCI card that can do LVD would do the job.
>
> Regards
> Olaf
>
> --
> Note: To sign off this list, send a "signoff networker" command via email
> to listserv AT listmail.temple DOT edu or visit the list's Web site at
> http://listmail.temple.edu/archives/networker.html where you can
> also view and post messages to the list.
> =*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
|
