[Users] Re: [Serusers] High Availability Setup

Greger V. Teigre greger at teigre.com
Wed Dec 21 07:06:38 CET 2005


Inline.

> Hi verybody,
>
> WARNING: Long e-mail.....
>
> I just started learning how to set up a "classical" configuration with
> load balancer, DB backend and so on....I was reading several threads
> of the mailing lists but since I am a begginer I would need some help
> and clarifications. Apologies if my questions have been solved before.
> [I read Greger and Granig (and more people, I guess) are working (?)
> on a document describing what I try to explain in the follofwing
> lines. I would like to know if you are on it or commercial strategies
> forced you to leave it...as it seems that HA setups are high
> confidential ;)]

Andreas is working on the setup and I have promised to replicate his setup 
and document it as an appendix/separate document from ONsip.org. The two 
main pieces of code (usrloc-cl and Path patch) are sort of stuck in between 
stable and head (usrloc-cl is for 0.9.x and must be ported to head, while 
the Path patch was made for head, but several changes to head requires a 
review/new patch).

> I'll try to draw what I have guessed from the threads and other 
> literature...
>
>                                          [SER1]
> [LB1] [SER2]
>      . [DB (mysql cluster)]
> [LB2]      .
> [SERn]
>
> LBi are the front-end load balancers.
> SER: do I have to say what are these?? (either openSER/SER. I will use
> the name SER for simplicity but I refer to both)
> DB: I placed mysql cluster because it seems to be the most common 
> backend...
>
> Let me make an overview...
>
> Load Balancer
> For LB, it seems that dispacther module can be used. (I don´t remember
> who proposed to dispatch REGISTER with To: and INIVTE with Req-URI,
> why not using Call-ID?? For distributing the registration of a given
> user between several registrars?).

This way you will always send all registrations for a given user to the same 
server and INVITEs to that user to the same server. Using Call-Id, you would 
spread the requests across servers randomly.

> Dispatcher module does not have
> failover in case SERi fails so somebody propose to use lcr. Using lcr
> module provides failover but does not guarantee to balance to the
> right instance of SER so I think the only option is to use dispatcher.
> Alternatives:
> Please correct me if I am wrong but there isn´t yet an LVR
> implementation being Call-ID aware....

That's correct, except commercial ones.

> DNS SRV records...lack of implementation in end-points and not 
> NAT-friendly..

You will probably want to use DNS SRV as well. However, implementation can 
differe between UAs.

> In order to provide high availability, I tried linux-ha between the 2
> LBi for simplicity although it has some wasted resources. Is it worth
> to give a try to other IP failover aplications (anycast, quagga/bgp,
> vrrpd)?

Well, the same issues are there. Also, there are some issues related to 
failover time. Linux HA is probably as good as any.  What you could do is to 
run two instances of SER on each server and let both load balance each other 
(using xen or vmware). The problem is that you still cannot put more load on 
both servers than what one server can take, in case one physical server goes 
down.

> Anyone with more experience can comment their use??
>
>
> SIP proxy
> SER....just a matter of taste what you want to use, SER or openSER....
> I read about the cacheless usrloc patch but I could not find it on
> CVS...has it been integrated into the head branch??? has it been
> deprecated because of ...(please fill in the reason)?

Can be found in experimental module of rel_0_9_0 branch.
g-)
>
>
> DB Backend
> For the DB backend...I am going to comment mysql clustering...I am
> using now 4.1.11 because it was already installed. Is it recommended
> to upgrade?? I tried setting the management node, a couple of data
> nodes, and I was trying to set a linux-ha configuration with ip take
> over for the mysql api node (is it possible??) but unsuccessfully:
> When the preferred mysql api node falls down, SER detects the
> connection is closed and all further requests are replied with
> "Internal Server Error" although the secondary mysql api has overtaken
> the service IP and ser database is properly populated.
> Is it possible to make SER reconnecting on mysql failure??  It
> reconnects to the preferred server once it comes back to live but it
> is impossible to connect to the secondary....
> As you might have guessed I am a total begginer with mySQL...
>
> Thank for reading until this line, all kinds of comments are wellcome,
>
> Samuel.
>
> _______________________________________________
> Serusers mailing list
> Serusers at iptel.org
> http://mail.iptel.org/mailman/listinfo/serusers
>
> 





More information about the Users mailing list