Hi Adam
Answers appear below.
tl
-----Original Message-----
From: Adam [mailto:adam-s AT pacbell DOT net]
Sent: Monday, March 17, 2003 4:18 PM
To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
Subject: [Networker] NDMP of netapp filling up /nsr/tmp
Hello,
We use Legato Networker 6.1.2 build 340 on a Solaris 2.8 box to backup a
couple of Network Appliance F802's via NDMP (as well as a bunch of Unix
servers). The unix boxes are backed up in different pools to a DLT8000
locally attached to the Sun Ultra 60 while the two Netapp's backup via NDMP
to a DLT8000 locally attached to netapp1. The backup for netapp1 is
running fine on Saturday, but the Sunday night backup of netapp2 keeps
failing because /nsr fills up and everything freezes/aborts. netapp1 has
68GB of data to backup while netapp2 (which is failing) has 128GB of data.
Each Netapp/client is configured identically on the Legato server with a
backup command of "nsrndmp_save -T dump", Application information:
"DIRECT=y, HIST=y, and UPDATE=y"; and Savesets for both are: /vol/vol0 and
/vol/vol1. Networker indexes are on a dedicated (25GB, 20GB of which is in
use) filesystem of their own (/nsr/index) so /nsr/tmp is on the 475MB /nsr
filesystem which only has 113mb free. I gave all the space to /nsr/indexes
because, other than /nsr/logs, little previously grew in size other than the
indexes. In fact, we've been backing up these 2 NetApps this for quite some
time and this has never happened until a week ago.
What I see, when I run something like sudo savegrp -n -l full -r 2 -c
netapp1 'NETAPP1'
is that scratch/temp looking files are created in /nsr/tmp. This happens
for *either* netapp so I don't think the fact the DLT is on the one not
failing is the problem (but I really don't know yet). Here is an example of
the files created by the above command:
-rw-r--r-- 1 root other 20314208 Mar 17 11:24 2fhdir21196
-rw-r--r-- 1 root other 1394100 Mar 17 11:25 2fhnode21196
What I am trying to find out is:
1. I am assuming that Legato buffers to these files because the network
(Gigabit) is faster than the tape can handle and thus turning off this
intermediate file write is impossible and/or not-adivsable. If those
assumptions are wrong, please correct me.
tl> NDMP servers generate file history information that is sent to the NDMP
client (NetWorker, in this case). NetWorker stores this information in the
files you see above. If the backup completes successfully, the NetWorker
will read in these temporary files, and write a second set of temporary
files. Then, it will read in this second set of temporary files, and write
that information to the client file indices in /nsr/index. So, you'll need
to have sufficient space in /nsr/tmp to handle 2x the largest set of files
you see above.
2. Is it possible to redirect the scratch/buffer file somewhere else from
within Legato or should I just link /nsr/tmp to another filesystem?
tl> I haven't tried this but I think that a link should work.
3. Since /nsr/index has all the space, would it harm performance horribly if
I linked /nsr/tmp there? /nsr/index is just a a few mirrored disks in a
diskpack so disk throughput is already not the best, but so far it has been
sufficient for our overnight coldbackups shop.
tl> Given that, in the final update phase, NetWorker will read from
/nsr/tmp and write to /nsr/index, it would certainly be better
(performance-wise) to have those two areas on different disks. But, unless
your environment is truly huge, a single well-performing disk should work
fine.
4. What are the sizing requirements for thes scratch/buffer files so I can
properly plan for growth? I.e. something like "Legato uses 1.5K per MB"
(which is what it would be if netapp1's 68GB was just filling up the 113MB
free.
tl> Check out Chapter 13 of the NetWorker Administrator's guide. You'll
find recommendations there for memory requirements (I requested these be
documented after having issues during NetWorker V6.0 testing). I don't
remember if disk space requirements are given. If they aren't, please let
us know, and one of us will ask for these to be included in the
documentation.
5. Is there a more thorough source of info/documentation that I missed?
The main document I used when I set all this up many months ago (and it has
been working fine, done restores and all that) is the Legato Networker,
Release 6.1 Unix Version Administrator's guide. If there is a better
document covering setup/tweaking, etc. then I missed it and would appreciate
a pointer to that as well.
tl> I know I read a good white paper on the NDMP implementation several
years ago, but this may have been an engineering-only document. I think the
NDMP chapter in the Administrator's Guide is good. If you've found
topics/problems that aren't covered, please let us know, and someone will
agitate to have them added.
My apologies if all this was alread covered, I checked the archives, google,
and websupport and found no addition information. So if I missed a thread
or subject that would get me to the correct information, please redirect me.
thanks for any info, background, pointers, etc.
Adam
--
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.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
|
