Messages in this thread | | | From | (Bret Martin) | Subject | Re: Backup to Exabyte aborts (1.3.??) | Date | Tue, 09 Apr 1996 11:28:51 -0400 |
| |
On Tue, 09 Apr 1996 21:38:44 +0930 "Stephen Davies" wrote: > For some time now I have been having apparently random failures of my > file system backups to tape. > > Initially I thought that it might have been a media problem as I got > better results if I backed up one disk at a time rather than doing > all three to one tape but lately (1.3.78 - 1.3.84) I get failures on > even quite small backups. > > The failures only seem to happen while backing up my scsi disk; I can > backup more than 450 Mb from IDE drives but now cannot backup 300Mb > from a scsi disk.
I suspect your IDE transfers to the Exabyte don't fail at all, correct?
I had this problem as well, and it seems that the Exabyte 8200 is somehow particularly ungraceful when *something* happens on the SCSI bus. I'm guessing from reading a little about the 8200 that it doesn't support the "disconnect" instruction of SCSI2 or something like that (I don't know much about SCSI at all, sorry if I'm sounding silly here)... In any case, whenever I tried dump/afio/tar/whatever *to* the tape *from* my SCSI-mounted Seagate drive, I had the exact same behavior as you, the SCSI bus was continually reset. From my EIDE drive, however, I had no problems. (My friend joked that perhaps the tape drive just couldn't keep up with the Barracuda drive, but the EIDE was slow enough to make it happy :-) )
In my case, it brought the machine down. I'm using an NCR53c810 card as the host adapter.
My solution, which I think is rather elegant at this stage :-), was to install my spare Adaptec 1520 as a second host adapter, and attach the Exabyte drive to that. So I now have two SCSI busses -- the one with all the disks on in on the NCR card and then the one the Exabyte has all to itself.
I am pretty certain this is not a kernel problem, but some problem with the firmware of the Exabyte -- I haven't investigated if there is an upgrade available, but my current solution is working quite well.
--Bret
-- Bret Andrew Martin bam@yale.edu
| |