Re: [CentOS] simple load balancing/failover for OWA

2008-05-28 Thread Amos Shapira
On Fri, Apr 4, 2008 at 11:36 PM, Matt Shields [EMAIL PROTECTED] wrote:
 On Fri, Apr 4, 2008 at 9:03 AM, Ruslan Sivak [EMAIL PROTECTED] wrote:
 David Hrbác wrote:

  Ruslan Sivak napsal(a):
 
   We are building an exchange cluster with two front end Outlook Web
 Access servers.  We would like to at least have some sort of failover, and
 prefereably load balancing for them.
  
   Russ
  
 
  Russ,
  take a look at
 http://httpd.apache.org/docs/2.2/mod/mod_proxy_balancer.html
  David
  ___
 
  Yes, that what I was thinking.  However, I believe these servers also run
 smtp, pop3, imap,etc, so I don't think Apache can handle all of those..

 LVS handles all protocols.  It can do any port and UDP or TCP.  It
 supports different types of balancing Round Robin (rr), Weighted Round
 Robin (wrr), Least Connections (lc), Weighted Least Connections (wlc).
  It can do sticky sessions, so if OWA doesn't like when you go from
 server 1 to server 2, LVS will keep the user stuck to one server.
 Plus a ton of other features.  Give it a shot.

I'm trying to setup a couple of LVS servers on CentOS 5 and hitting
difficulties.
I've been googl'ing around the docs (it mostly all gets back to
http://www.austintek.com/LVS/LVS-HOWTO/mini-HOWTO/LVS-mini-HOWTO.html)
for a few days now but can't get it to work. It sometimes work
temporarily but very quickly a few fail-overs made things totally
unworkable.

One issue I suspect that might be in my way to success is that the two
servers (I'm trying to use same two servers both as ldirectors and
real servers) are assigned to two separate /28 networks, BUT the
virtual rack (from ThePlanet) allows both of them to be assigned
addresses from each other's network so I assigned an address from the
correct network on an alias on the other server (eth0:0).

Does anyone have a working example for this on CentOS 5? The docs I
found so far are mostly generic and talk a lot about ancient kernel
versions like 2.0, 2.2 and 2.4.

I subscribed to the lvs-users mailing list but I'm not sure how active it is.

Thanks,

--Amos
___
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos


Re: [CentOS] simple load balancing/failover for OWA

2008-04-04 Thread Matt Shields
On Fri, Apr 4, 2008 at 9:03 AM, Ruslan Sivak [EMAIL PROTECTED] wrote:
 David Hrbác wrote:

  Ruslan Sivak napsal(a):
 
   We are building an exchange cluster with two front end Outlook Web
 Access servers.  We would like to at least have some sort of failover, and
 prefereably load balancing for them.
  
   Russ
  
 
  Russ,
  take a look at
 http://httpd.apache.org/docs/2.2/mod/mod_proxy_balancer.html
  David
  ___
 
  Yes, that what I was thinking.  However, I believe these servers also run
 smtp, pop3, imap,etc, so I don't think Apache can handle all of those..

LVS handles all protocols.  It can do any port and UDP or TCP.  It
supports different types of balancing Round Robin (rr), Weighted Round
Robin (wrr), Least Connections (lc), Weighted Least Connections (wlc).
 It can do sticky sessions, so if OWA doesn't like when you go from
server 1 to server 2, LVS will keep the user stuck to one server.
Plus a ton of other features.  Give it a shot.


-- 
-matt
___
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos


[CentOS] simple load balancing/failover for OWA

2008-04-03 Thread Ruslan Sivak
We are building an exchange cluster with two front end Outlook Web 
Access servers.  We would like to at least have some sort of failover, 
and prefereably load balancing for them.


The MS recommended way is to use NLB, but for various reasons that's not 
working with our set up. 

We are looking to set up a single linux server and use something like 
LVS to load balance/fail over the connections. 

Looking at LVS, it looks like it hasn't been updated in a while.  Is it 
stable?  Is it still the preferred solution?  Since OWA has sessions, 
but no session replication, we would need something that can keep 
directing the same client to the same server (either by IP, or 
preferably by cookie).  Can LVS handle this?


I also saw that the latest version of Apache has some sort of load 
balancing support.  Can it be used to set something like this up?  We 
are looking for the simplest solution that won't require a lot of 
maintanence.


We understand that having a single LVS box creates a single point of 
failure, and are willing to accept the risk for now to keep things 
simpler. 


Russ


___
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos


Re: [CentOS] simple load balancing/failover for OWA

2008-04-03 Thread Matt Shields
On Thu, Apr 3, 2008 at 11:07 PM, Ruslan Sivak [EMAIL PROTECTED] wrote:
 We are building an exchange cluster with two front end Outlook Web Access
 servers.  We would like to at least have some sort of failover, and
 prefereably load balancing for them.

  The MS recommended way is to use NLB, but for various reasons that's not
 working with our set up.
  We are looking to set up a single linux server and use something like LVS
 to load balance/fail over the connections.
  Looking at LVS, it looks like it hasn't been updated in a while.  Is it
 stable?  Is it still the preferred solution?  Since OWA has sessions, but no
 session replication, we would need something that can keep directing the
 same client to the same server (either by IP, or preferably by cookie).  Can
 LVS handle this?

  I also saw that the latest version of Apache has some sort of load
 balancing support.  Can it be used to set something like this up?  We are
 looking for the simplest solution that won't require a lot of maintanence.

  We understand that having a single LVS box creates a single point of
 failure, and are willing to accept the risk for now to keep things simpler.
  Russ

LVS is extremely stable.  We have 3 active/passive clusters of LVS
servers, for different projects, and they work extremely well.

The way we set it up was each LVS pair is running heartbeat,
ldirectord and ipvsadm.  If the primary node goes offline for any
reason, heartbeat on the second tells it to take over.  And there are
lots of config settings to tweak heartbeat.

-- 
-matt
___
CentOS mailing list
CentOS@centos.org
http://lists.centos.org/mailman/listinfo/centos