[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
OID comression tests.
We did some experiments of OID compression which may be of
interest to the working group.
To test our management applications we simulate the management of huge
networks where every computer
represents the SNMP agents of 100's of network devices.
It turns out that memory usage is the main bottleneck. Therefore, our
goal was to reduce the memory footprint of a MIB.
We wanted to check the possibility of assigning a static 32 bit
tag to every OID type in every MIB module, and use tag.instance in
all our internal representations of OID's. Translation from
OID_type.instanace to tag.instance and back would be done only
when necessary, using a static mapping file. To check if this
solution actually reduces memory consumption we recorded the MIB
of a core FORE ATM switch (about 10Mb of data in OID,value
format). We then took all the MIB modules we had for the device
and assigned tags for the OID types of all the tables. Finally, we
replaced the prefix of the OID of every instance of these tables in the
recoded data with the tag. The result of this operation reduced
the file size by about 1/2 (~5Mb).
These results sort of surprised us because we assumed the instance
part of the OID of most huge tables will be much smaller than the
object type (usually these core devices do not support fancy MIBS).
We didn't tag OID's that were part of the data so perhaps this
also adds to the size of the file. Still, 50% reduction is not bad.
Hope this helps,