Mailing List Archive: multiple virtual services on real servers

Hi,
I'm currently using the Ultra-Monkey package from VA Linux which is
basically LVS packaged with the HeartBeat package for high availability on
the load balancer end.

I've been running a single service ( http ) for several months now with
great success. The service is balanced across 5 real web servers using
direct routing. Yesterday, I added a second and third service for another
http LVS and an https LVS. Both of the services I added yesterday are on
the same IP address.

The new services are working properly but they seem to be significantly
slower than my original service ( which is still quite fast after adding the
2 additional services ). The only difference between the original service
and the new one is that the original service had 5 real servers with private
IP's ( RFC 1918) and each private IP was statically linked to a publically
addressable IP. The services that I added yesterday are also using 5
private IP's but none are statically linked, and they are all using a single
PAT address for outbound traffic. I have correctly added a second dummy
alias on each of the real servers. ( now there are two dummy aliases lo:0
and lo:1 ).

My question is: Am I seeing a performance hit because I'm trying to use the
'real' servers to service more than one virtual service, or is this because
of the statically linked IP's, or does it seem like something entirely
different?

It seems to me that if the performance problem was a result of trying to run
multiple virtual services on a each real server, then my original virtual
service would have slowed down as well.

If any of what I said above was unclear, let me know, and I'll try to better
explain what is going on.

--Ben

----------------------------------------------------------------------------
--
Ben Nelson
bnelson@rightnow.com

Hosting Systems Administrator
RightNow Technologies
Customer Service For Every Web Site

Hi Ben,

I'd suspect that since you're using non RFC-1918 IPs, that these 'real'
servers might not be on the same local network/switch as your other
ones.

Have you measured latency from the LVS server to the new real servers
and compared it with the old RFC-1918 servers?

Might seem obvious, but that's the only thing I can tell right now.

Send me your lvs.conf file, I might be able to find something in there.

There are no scalability issues that you should be running into in this
case, so I think you're barking up the wrong tree there.,

By the way, I noticed that you guys are hosted with Exodus. I don't mean
for this to turn into a sales pitch, but can we open a conversation for
hosting?

We are the only provider to commercially _support_ LVS (and support the
free LVS project, check out the sponsors page). We host Porsches 210
North American Dealers.

I'd be very interested in the possibility of earning rightnow.com's
business.

Cheers,
Raj Dutt
Voxel dot Net, Inc

"Nelson, Ben" wrote:
>
> Hi,
> I'm currently using the Ultra-Monkey package from VA Linux which is
> basically LVS packaged with the HeartBeat package for high availability on
> the load balancer end.
>
> I've been running a single service ( http ) for several months now with
> great success. The service is balanced across 5 real web servers using
> direct routing. Yesterday, I added a second and third service for another
> http LVS and an https LVS. Both of the services I added yesterday are on
> the same IP address.
>
> The new services are working properly but they seem to be significantly
> slower than my original service ( which is still quite fast after adding the
> 2 additional services ). The only difference between the original service
> and the new one is that the original service had 5 real servers with private
> IP's ( RFC 1918) and each private IP was statically linked to a publically
> addressable IP. The services that I added yesterday are also using 5
> private IP's but none are statically linked, and they are all using a single
> PAT address for outbound traffic. I have correctly added a second dummy
> alias on each of the real servers. ( now there are two dummy aliases lo:0
> and lo:1 ).
>
> My question is: Am I seeing a performance hit because I'm trying to use the
> 'real' servers to service more than one virtual service, or is this because
> of the statically linked IP's, or does it seem like something entirely
> different?
>
> It seems to me that if the performance problem was a result of trying to run
> multiple virtual services on a each real server, then my original virtual
> service would have slowed down as well.
>
> If any of what I said above was unclear, let me know, and I'll try to better
> explain what is going on.
>
> --Ben
>
> ----------------------------------------------------------------------------
> --
> Ben Nelson
> bnelson@rightnow.com
>
> Hosting Systems Administrator
> RightNow Technologies
> Customer Service For Every Web Site
>
> _______________________________________________
> LinuxVirtualServer.org mailing list - lvs-users@LinuxVirtualServer.org
> Send requests to lvs-users-request@LinuxVirtualServer.org
> or go to http://www.in-addr.de/mailman/listinfo/lvs-users