At 01:04 AM 10/5/2004, Michael Shuler wrote:
My intent is to do SIP load balancing using a layer 4 hardware switch (such as a Foundry ServerIron XL) or a layer 4 software switch (such as UltraMonkey).
Since SER *could* operate in a completely stateless mode it could serve as a per packet proxy in front of a series of stateful feature servers such as Asterisk. A L4 device by itself would not be enough and a stateless MUST be in front to do the actually balancing because of SIP's flaw/design of containing routing information throughout the packets life.
I don't think that this specific issue is SIP design flaw. It is coming from the proxy mode which is useful for many reasons. Achieving consistency with a proxy and load-balancer is not as easy as for banal request-response apps like web.
It would have been nicer if it only carried a source and destination and worked more like TCP/IP but unfortunately it is what it is.
Anyway, I about have everything working and I only seem to have a small issue when the second SER server comes online with some strange message passing between them. Most of it I believe is caused by my lack of understanding of how SER makes some decisions on where to send things.
It depends on what you mean by things. If you mean log error than these are sent to syslog. If you mean replies, their destination is governed by RFC3261 and rport extensions. If you mean proxied requests, than the destination is governed by routing script.
I think what would help is if I setup a "work in progress" web site that would show all my config files and the layout of the whole thing along with an explanation of why I decided to do what I did. I should have it up later today or tomorrow.
Michael Shuler, C.E.O. BitWise Communications, Inc. (CLEC) And BitWise Systems, Inc. (ISP) 682 High Point Lane East Peoria, IL 61611 Office: (217) 585-0357 Cell: (309) 657-6365 Fax: (309) 213-3500 E-Mail: mike@bwsys.net Customer Service: (877) 976-0711
-----Original Message----- From: Jiri Kuthan [mailto:jiri@iptel.org] Sent: Monday, October 04, 2004 5:42 PM To: Michael Shuler; serusers@lists.iptel.org Cc: jan@iptel.org Subject: RE: [Serusers] forward() and t_relay() differences
At 07:50 PM 10/4/2004, Michael Shuler wrote:
Sorry if that first line sounded snotty, I didn't mean it
that way. I
didn't read it until after I sent it. What I meant to say
was thank you for
the response.
you are very welcome.
I had already been through the docs though and found that too but I was still seeing the following problem..
I am interestd in more feedback on load-balancers. To my knowledge, it is a technology which has some conflicts with SIP protocol and those LBs that try to fix the problems using built-in SIP awareness don't do necessarily any better. We are working on a SER built-in load-balancing architecture but that's still work in progress.
-jiri
-- Jiri Kuthan http://iptel.org/~jiri/
-
Jiri Kuthan