INTERIM_MEETING_REPORT_
Reported by Bob Purvy/Oracle and David Brower/INGRES
Minutes of the Relational Database Management Systems MIB Working Group
(RDBMSMIB)
Agenda
o Discuss Mike Hartstein's proposal for an ``analyst's briefing'' in
April, when the group's work has been completed.
o Introductions and chairman's review of what has gone on in the
group lately (Microsoft invitation, IETF sessions, etc.).
o Review latest Internet-Drafts. The goal is to settle all
outstanding issues with the MIB as it is now, before addressing new
functionality. While it is not definate that the group can do it
at this meeting, there is some chance of it.
Outstanding Issues
o Use of Network Services MIB. What are the implications? This may
take awhile!
o rdbmsSrvInfoUpTime and other time-related issues. When do the
statistics have to get reset?
o Marc Sinykin's issues (message of 20 February):
rdbmsSrvInfoDiskOutOfSpaces, etc. General question of the proper
handling of variables which cannot be gotten by the database
(sub)agent.
o Whatever else people come up with. DEC is sending someone this
time, so there may be some new issues from them, or from other new
attendees.
New Issues
o Prioritize the new features that have been proposed, or that are
proposed at the meeting. Which are the most critical to get in?
The goal is just to touch lightly on the issues, and make sure that
someone is tasked with coming up with a concrete proposal on the
group's top priorities.
Discussion
o Some members wanted more discussion of the ``analyst's briefing''
proposal; this will be carried out via e-mail.
o It had appeared earlier that very few members were planning on
attending the Seattle IETF, but it looks like quite a few are now
registering or planning to, so we should have a good turnout.
o The question was brought up about the difficulty of computing
``space used''; several vendors find this extremely costly to get.
It was pointed out that your implementation can cache the result
and refuse to compute it more often than you find acceptable. We
also decided to insert the word ``estimated'' in the description
for that variable.
o The Network Services MIB (RFC 1565) was discussed at length. No
one much liked the implementation difficulties; using a subagent
technology, it implies that every network service on the host must
choose a unique index via some method which is outside the SNMP
standard. It was remarked that some master agents are capable of
accepting registrations for a particular row of a conceptual table,
and/or assigning a unique index themselves, while others cannot do
either.
Various alternatives were debated. Ultimately we decided that it
was ``the right thing to do,'' and furthermore, not supporting
RFC 1565 would likely cause problems getting our standard approved.
There will be difficulties for some period of time as hardware
vendors and SNMP agent providers come up to speed with the new
requirements of application monitoring, but we really do not see a
good alternative to doing this.
Dave Brower had already circulated a draft MIB with RFC 1565
incorporated; this will be refined on the basis of what was
decided, and reissued as an Internet-Draft.
o The question arose on how to denote ``unlimited'' in the
MaxSessions variable. Some standards groups have used zero in
cases like this, and others have used -1. We choose zero.
o The questions about counters and ``up time'' were debated.
Marshall explained the SNMP philosophy of polling and counters
(basically, only the change in a counter's value is to be
considered significant).
(This discussion assumes a master agent/subagent architecture,
where the same subagent implements its rows in the Network Services
MIB and in the RDBMS MIB.)
`applUptime' will assume a new value any time the subagent/RDBMS
changes state. This is the signal to the management application
that a sampling discontinuity may have taken place. Therefore, the
drill is when the subagent comes up, it gets the value of
sysUptime, and makes that the applUptime. If the value of
sysUptime is unavailable, it uses zero.
o rdbmsRelActiveTime - remove the word `first.'
o Stan Wong's message was reviewed. We agreed to change the wording
of ``vendor'' to allow for multiple products from the same vendor
(``vendor/product'' or some other such choice of the editor)
We accepted Stan's proposal to remove rdbmsDbInfoCreateTime from
the MIB, since it does not add much to manageability.
``Time of last backup'' is to be added to the wish list.
o A paragraph should be added to the overview explaining that this
MIB is not intended to be sufficient in and of itself for
performance tuning an RDBMS, and that the vendor's private MIB, if
it exists, will be a more reliable guide.
o The wording on DbInfoTable was debated at some length. Oracle felt
it was important that the wording not preclude a row being present
if a database is not active. The upshot is that Marc Sinykin, Dave
Brower, Marshall Rose, and Bob Purvy will confer electronically and
propose a wording.
o A ``contact person'' variable is to be added to the Server table,
analogous to rdbmsDbContact.
o For ``disk reads'' and ``disk writes,'' the word ``physical'' is to
be removed. We cannot know that a read or write issued to the
operating system actually resulted in a physical read or write.
(Note that this variable does not count reads or writes which may
be satisfied out of the vendor's own cache; this is still to count
only reads or writes issued to the operating system, or disk
subsystem if no OS is being used.)
o Stan Wong will try to find a definition of ``database'' acceptable
to all. We wish him luck in this endeavor.
o There is some sentiment to remove
rdbmsSrvInfoUncommittedTransactions from the MIB. No consensus was
reached yet, but for now a comment will be added that this variable
is controversial.
Attendees
David Brower daveb@ingres.com
David Campbell campbell@nova.enet.dec.com
Anthony Daniel anthony@informix.com
Craig De Noce craigd@sybase.co
Howard Dernehl howard@ingres.com
Bill Kehoe kehoe@sybase.com
Brian Promes bpromes@novell.com
Robert Purvy bpurvy@us.oracle.com
Marshall T. Rose mrose.iesg@dbc.mtview.ca.us
Marc Sinykin msinykin@us.oracle.com
Jay Smith jaysmith@us.oracle.com
Stan Wong swong@vnet.ibm.com