Yes, we are. At first, we too wanted to keep x days worth of data in the
collection binary files. Our manuals have an example on how to do this but it
contained many syntax errors. We had to play with it a bit to get it to work.
However, once we got it working we discovered two things. The first is that the
binary reload process was way too slow for the amount of data we were pushing
through it. The second thing is that keeping anymore that one days worth of data
in the binary files was too much data to effectively see in the graphs anyway
(we have 58,000+ objects in our database.)
So, now we dump the binary collection files every day in the early morning
hours, then delete them.
This effectively assures we keep no more than 24 hours worth of collection data
in the binary files.
After dumping the binary files, we reformat the ASCII output to our liking, then
FTP to MVS/DB2 which does an automatic DB load as soon as the FTP x-fer is done.
We collect Cisco's avgBusy5 CPU load MIB and Input/Output line utilization via
MIB expressions every 15 mins on nearly 800 routers (some of which have over 140
interfaces.) The entire process takes anywhere from 3-5 hours to run depending
on network traffic (we make additional SNMP calls during the running of
snmpColDump). Every 24 hours, we dump, re-format, FTP and load into MVS/DB2,
over 140 MB's of collection data! The database purges data older than 60 days.
The process finishes several hours before our help desk staff arrive and is
remarkably reliable.
Now if we could only find a graphing/trending tool like Concord's Network Health
that didn't cost an arm and a leg, and could use our collected data as input, we
would be set. Our customers are demanding rather loudly, performance/trending
stats on how their networks are performing.
I can email you privately if you want to talk specifics.
Brook, Bryan S wrote:
> Anyone here sending their Netview collections to a RDBMS? We are
> and want to store only the last 2 weeks worth of data. So we want to script
> up "delete collection data > 2 weeks old" . How do you get a "- 2 weeks"
> with the date format as shown? I can't believe this isnt a common request,
> however the command wont support this without some prior math manipulation.
> Anyone have it down already?
>
> Thanks,
> Bryan
|