On Fri, 2017-06-30 at 15:30 -0400, Greg Dickie wrote: > Hi Alan, > > Thanks for the response. It looks to me like taper is slow. I'm > trying to not use a holding disk, the NAS is 250TB and doesn't make > sense to have 250TB locally to cache it. It's connected by a 10G > ethernet connection so I "should" be able to stream pretty quickly > from it. i did put a 1TB disk as a part cache but that seemed to > behave the same way. > > I cranked up the debug level for taper and this is what I see > repeatedly: > Fri Jun 30 15:24:16.549381220 2017: pid 23980: thd-0x2399850: taper: > XDTS: writing 2097152 bytes to device > Fri Jun 30 15:24:16.762549619 2017: pid 23980: thd-0x2399850: taper: > XDTS: writing 2097152 bytes to device > Fri Jun 30 15:24:17.214824233 2017: pid 23980: thd-0x2399850: taper: > XDTS: writing 2097152 bytes to device > > > which is just writing the 2M blocks I guess. > > I'm new to amanda so it's likely I just am confused about how it > should be set up. > > What information can I provide that would help narrow it down? > >
I haven't had the pleasure of using hardware that big or fast. Logically, though, if it was working fine before, and if the tape drive is still working correctly (test with dd?), and if the only settings you've changed are adding part_size and part_cache_* settings, then ... I would think the bottleneck is probably wherever you're writing/reading parts to. 300MB+/sec writes and simultaneous reads are no joke. If your part_cache is in RAM, though, then I'm out of ideas. Do make sure you didn't enable software compression along the way.
