GREAT information! Thank you very much.
That gives me something to shoot for. Unfortunately we are saying goodbye to
NetView by the end of the year so I just need to get things running better for
now until we sunset. Sure has been a good run though. Overall I would say that NetView
has been a great toolset.
Thanks again for your input and advice!
From:
owner-nv-l@lists.us.ibm.com [mailto:owner-nv-l@lists.us.ibm.com] On Behalf Of Stephen Hochstetler
Sent: Wednesday, March 15, 2006 11:40 AM
To: nv-l@lists.us.ibm.com
Subject: RE: [nv-l] Ping is falling behind
Scott,
Time flies, it has been a full 10 years since I have 'optimized' a big NetView
box. The underlying technology probably has not changed that much. There are a
couple of key ways to lower your CPU usage.
1) Instead of using a golden /etc/hosts file with an entry for each 'management
interface', put a DNS on your NetView server and get those 1300 interfaces in
there for both name and reverse lookup. You will see a difference. For every
function NetView is reading that file, one line at a time, over and over, 1000s
of times per 5 minutes.
2) Build a reverse lookup table for your 38K interfaces. Best way to do that is
to write a script that does an ovtopodump and generates these files for local
DNS. You will see hugh difference
When I did those two things about 10 years ago I saw a large 4-way AIX box go
from 70% CPU usage to 6% CPU usage, your milage may vary.
Once I had the server humming along then the user's noticed a lag about once a
minute in responses to their maps. After investigation I found that it was how
ovwdb and AIX interacts. You have a large cache size for ovwdb right? This
means that the whole ovwdb database is in memory. AIX wants to make sure what
is in memory also matches what is on the hard disks, so he has a routine that
dumps the application caches to hard disk every 60 seconds. (at least AIX 4.2
and 5.1 did). While AIX does that he halts the ovwdb daemon to ensure integrity
with what is being written, which halts any other daemon including maps or
netmon while that data dumps. This was due to my /ovw filesystem being stored
on just 1 disk, I was seeing a delay due to 150MB being written once a minute
onto one spindle. The fix was to simply create a file system that was stripped
across 3 hard drives and mount it so that the ovwdb database was stored there.
Stephen Hochstetler shochste@us.ibm.com
International Technical Support Organization at IBM
Office - 512-838-6198 (t/l 678) FAX - 512-838-6931
http://www.redbooks.ibm.com