Hey all,
We are implementing a modified version of draft-ietf-sip-mib-09.txt,
where we've stripped out the use of the NETWRK-SERVICES-MIB and are
adding X.731 state information (similar to what is covered in RFC 4268)
and additional information we feel we need to capture to successfully
manage an openSER (http://openser.org/) instance. For example, tracking
active phone calls is something we need that isn't supported by the
SIPMIBs.
Most of the data we need to expose is pretty straightforward and doesn't
require a lot of ingenuity. However, there are two tables in the
SIP-SERVER-MIB that are potential problems: sipRegUserTable and
sipContactTable. The sipRegUserTable has a list of all registrations on
the server (think one per SIP phone to make life easier). That list is
potentially unbounded, limited only by available memory and storage.
The sipContactTable can potentially be worse, as it represents contact
information per registration. A single registration can have multiple
contact rows.
The proposed communication mechanism with openSER is to use the official
FIF interface, where essentially we shove some text into the FIF and
we get a text response back. At the moment, we've decided to not add
AgentX support directly into openSER. So all of the data is going to
live in the openSER instance and we will be poking it via the FIF to
expose it in the SNMP agent. No problem so far. Now I need to manage
the sipRegUserTable, where I don't have the list of current
registrations in memory. Querying for that list (just the index
information, not the entire table row) may be prohibitive if it's
several thousand entries. I don't know at this time if the module that
tracks reservations will be our own, or one of the stock modules from
openSER. If it's our own, we can likely keep the list of registrations
in a sorted order; if it's not, we can't depend on the registrations
being properly sorted.
Any suggestions on what approach I should take? While the FAQ entries
for mib2c are helpful, and point me to the iterate or mfd config files
to deal with tables that have external data, I was hoping someone could
point out in advance whether or not the standard approaches will scale
to the 10,000+ row case.
I'll poke away at it myself and report back whatever progress I make.
Thanks,
Glenn McAllister
SMA Networks, Inc.
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems? Stop! Download the new AJAX search engine that makes
searching your log files as easy as surfing the web. DWNLAD SPLUNK!
Net-snmp-coders mailing list
Net-snmp-coders (AT) lists (DOT) sourceforge.net

Robert Story wrote:
Thu, 22 Dec 2005 09:40:45 -0500 Glenn wrote:
GMI don't know at this time if the module that
GMtracks reservations will be our own, or one of the stock modules from
GMopenSER. If it's our own, we can likely keep the list of registrations
GMin a sorted order; if it's not, we can't depend on the registrations
GMbeing properly sorted.
GM
GMAny suggestions on what approach I should take?

It really depends on whether or not your data is sorted, if it's sorted
properly, and if the fifo allows access to individual entries or just the
whole list.

I've been doing some digging and it turns out the usrloc module tracks
most of the information I need. I can write custom commands to get the
information in the format I need. Unfortunately, the current way the
module stores the data is in an unsorted linked list. I've got a custom
command that lists just the address of record (my index value, its a
(usually) short value like 4161234567 (AT) some (DOT) domain.com), but not in
sorted order. Looks like I'll have to extend my command. that note,
I'm not sure I know what you meant by "sorted correctly." I'm happy to
RTFM, if someone will point me at the FM. :-)

I have the indexes in sorted order, is there anything I need to do
to tell the cache that it's already sorted?

They do have a command that allows you to get an individual entry, but
it's not overly useful as it's missing just about everything we really
need, so I'll be writing a custom command to make up for it.

GMWhile the FAQ entries
GMfor mib2c are helpful, and point me to the iterate or mfd config files
GMto deal with tables that have external data, I was hoping someone could
GMpoint out in advance whether or not the standard approaches will scale
GMto the 10,000+ row case.

Is your system severely memory limited? If so, caching all the data could be an
issue and iterate is the way to go. But it will have *terrible* performance.
If memory is not so limited, MFD will allow you to work with indexes only,
until the rest of the data is needed.

This is a server box, not an embedded device, so while memory isn't
free, it's not nearly as scarce a resource as it might be. I'm going to
go ahead and work with the MFD approach (populate the index before the
data) for now. Performance is always a concern for us, so hopefully the
memory trade off will be acceptable.

Thanks,
Glenn McAllister
SMA Networks, Inc.

This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems? Stop! Download the new AJAX search engine that makes
searching your log files as easy as surfing the web. DWNLAD SPLUNK!

Net-snmp-coders mailing list
Net-snmp-coders (AT) lists (DOT) sourceforge.net

Robert Story wrote:
Fri, 06 Jan 2006 14:01:29 -0500 Glenn wrote:
GMI'm going to
GMgo ahead and work with the MFD approach (populate the index before the
GMdata) for now.

, if you are going with MFD, then you don't need to worry about trying to
get the list sorted, the container will keep itself sorted. This also means
you don't need to worry about what 'sorted correctly' means either. ;-)

Sweet. :-)

Do you have any idea how long it takes to get all the indexes from the server?

How long will almost certainly be a factor of how loaded the system is,
and on the server environment it might be *very* loaded (there is
another instance of SER that will run in the base station, but the
load there will be in the hundreds, not thousands; the server will
likely be aggregating services for all of the base stations).

That being said, the agent is likely the only thing querying the SER
server, so from a management perspective I should have the server all to
myself. Additionally, the data I need to manage will be relatively
static, as reservations have long durations (typically on the order of
hours). The only exception to this rule is the packet-storm scenario
when one or more sectors are restarted and all of the CPEs are trying to
re register themselves.

Since you do intended to be dealing with large amounts of data, if you can get
incremental updates from the server module (eg record add/delete), then you
wouldn't have to keep flushing/reloading the cache periodically.

I considered that approach, but it requires a lot more work in the
existing usrloc module. Currently it doesn't store any information
about when users are registered or unregistered. In particular, I'd
have to keep track of deleted records in a new list. It's not like any
of this is undoable, or even hard for that matter, it's just that I'm
running out of time. As always, the deadlines are shorter than the
amount of time needed to do the job right the first time

Thanks for the insight and input, it's appreciated.

Glenn McAllister
SMA Networks, Inc.

This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems? Stop! Download the new AJAX search engine that makes
searching your log files as easy as surfing the web. DWNLAD SPLUNK!

Net-snmp-coders mailing list
Net-snmp-coders (AT) lists (DOT) sourceforge.net