Some first thoughts

From: Kostas Anagnostakis <kostas@dont-contact.us>
Date: Wed, 19 Nov 1997 14:55:05 -0800 (PST)

Hi (and welcome to the list!),

The first release of squid w/ snmp support will be out late afternoon or
tomorrow, but here are some first thoughts/questions/issues, regarding
SNMP-based cache management.

- configuration management: it would surely be nice to be able to remotely
  tune cache's operational parameters, add/remove peers etc. However, I'm
  not sure whether this would be a good approach for security reasons,
  even using SNMPv2 (but I guess it's the lack of knowledge that makes me
  fear, anyobdy having experience with read-write SNMP or any other
  thoughts on that?)

- security management: modifying ACL's would be a great relief to
  some cache managers, who must keep per-user restrictions on cache
  access, and for large caches adding siblings etc. The above security
  considerations apply for this too.

- performance management: what parameters should be monitored? is per
  protocol/per relationship hitrate/bytecounts enough for evaluating
  performance? How deep should we go, given that we want to move
  computational burden to the manager's side and keep the cache's agent
  as simple as possible w/o having to transfer large data sets? are there
  any "base metrics"?

- accounting management: this one hurts a lot. It would be nice to have
  snmp based accounting, a la IETF's RTFM, but I'm afraid this will cost
  the cache too much in bulk transfers as well as in encoding [remember
  that even keeping access.log's is expensive how about snmp's encoding?].

Apparently 1) there are some things that should be left out of this
framework. 2) there should be convergence to a commonly supported
information model 3) we should work out how extensions for every specific
cache solution should be glued to the framework. 4) the relationship to
other MIB's (host-resources etc) also needs some thought: i'd rather keep
the cache MIB self-consistent for both practical reasons (some boxes have
really lousy snmpd implementations) and conceptual (we don't want
dedicated cache boxes having to implement each and every host-resources
field if we're still only interested in just cache statistics!)

The MIB published on cache-snmp home page, is rather squid-specific at
this time, but we'll be glad to wipe everything out/replace it with better
definitions.

Hope this opens some discussion... It's been a little bit
unstructured .. sorry for this raw braindumping ;-)

-Kostas
Received on Wed Nov 19 1997 - 14:55:19 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:21:59 MST