Correct me if I’m
wrong, but since SER is capable to handle tens or hundreds of thousands of SIP
messages per second, why not cut this number down to thousands and compose a HA
and Redundancy solution combining some stateless proxy in front of a bunch of
servers using cache based (like in 2 – MySQL cluster, p.ex.)? The stateless
proxy could have some local replication (two peers: one active and one
inactive) sharing a common IP (registered on SRV) for HA, and the geographic
issue could be achieved with a complete replication of this set.
The down-point in this
approach is that in a geographic switch there is no guarantee that the clusters
would be totally consistent between the two sets. The fall can (and Murphy says
it will) occur just before, or in the middle, a big replication DB operation. But
if You can accomplish that than I see this as a good solution.
Comments?
Edson.
From: serusers-bounces@lists.iptel.org
[mailto:serusers-bounces@lists.iptel.org] On
Behalf Of Greger V. Teigre
Sent: sexta-feira, 7 de julho de
2006 06:49
To: G.Jacobsen
Cc: seruser List
Subject: Re: [Serusers] Global
Failover Server
Yes, but this depends on your deployment setup and
policies. Obviously, you cannot support all features with all UAs. Three-way
conferencing for example. There's no guarantee that REFER and NOTIFYs work for
three random UAs from different vendors. There are still too many bugs around.
So, you need to say: If you have this and that UA, you will probably not get
three-way conferencing. Not much difference from saying: If you have this
and that UA, you may experience service outages.
There are several ways of creating server-side redundancy and scalability.
Unfortunately, none are trivial. There are three I have heard have been
successfully used in large-scale setups:
1. Cacheless usrloc with a mysql cluster as back-end DB combined with
implementation of the Path header (to find the registrar of a given UA). No
replication across servers
2. Multiple SER registrars, each with a standalone, local DB and where
SIP is used to replicate registrations. By storing replications from a
peer in a location_peer1 table and then lookup using this table, you can route
INVITEs to the registrar being able to pinhole the NAT in front of a given UA
3. Each SER is connected to a single mysql DB cluster as in #1, but since
usrloc also is in memory (cacheless usrloc is not used), replication is done
between the SER servers and save_memory() is used to store the location only in
memory (the registrar updates the cluster with save())
Each of these three can be combined with either:
a. call-id sticky front-end load balancer (commercial or modified LVS)
b. DNS SRV
c. Linux HA creating two and two peers
Only b) combined with either 1) or 3) or 2+c) can give geographic, client and
server-side redundancy and scalability.
Scalability and redundancy is sort of a pet project of mine... I would like to
see a simple, clear-documented and code-supported setup that satisfies most
common requirements. Currently I'm leaning against #2 (which actually can be
implemented in vanilla SER 0.9.x, but would be easier and better with built-in
support). I'm thinking about a setup where two and two servers are peers to eachother
using Linux HA. Both servers would have one active SER instance and one
inactive ready for taking over for the peer.
I have posted this overview to:
http://www.iptel.org/drupal/failover_redundancy_and_scalability_overview
See also:
http://www.iptel.org/drupal/ser/wishlist
g-)
G.Jacobsen wrote:
Samuel,
Do you happen to know what percentage of UAs out there
are really "Compliant" UAs ?
My impression so far regarding SRV DNS
records is that they are theoretically a nice feature but unfortunately almost
useless since one needs to cater for those non-compliant UAs anyway. I would
love to be convinced of the contrary.
Can anyone supply real usage figures
regarding compliant/non-compliant UAS ?
TIA
Gerry