Jeffrey A Schoolcraft wrote:
>
> I'm curious if there are any known DR LVS bottlenecks?
Even with perfect LVS code on the director, a VS-DR
director's tcpip layer will be saturated by a single
real-server with the same hardware. To get more
throughput, more directors are needed.
With the current failover method used by Ultramonkey,
one director is on standby all the time, ie during normal
operation, only half of your director capacity is being
used. As well, on failover all your connections are lost.
It would be better to have both directors working in
normal operation and during failover, to switch the
dead one out. This way, directors are running at
half capacity only during failover.
One way of doing this would be to have two IPs for the
site in the DNS tables, and for each director to serve
up the content through one of the VIPs (the real-servers will
have 2 VIPs). During failover, the up director assumes
the extra VIP and only half the connections are lost.
Presumably if you use 5 directors, then only 20% of the
connections are lost on failover.
Joe
--
Joseph Mack PhD, Senior Systems Engineer, Lockheed Martin
contractor to the National Environmental Supercomputer Center,
mailto:mack.joseph@epa.gov ph# 919-541-0007, RTP, NC, USA
>
> I'm curious if there are any known DR LVS bottlenecks?
Even with perfect LVS code on the director, a VS-DR
director's tcpip layer will be saturated by a single
real-server with the same hardware. To get more
throughput, more directors are needed.
With the current failover method used by Ultramonkey,
one director is on standby all the time, ie during normal
operation, only half of your director capacity is being
used. As well, on failover all your connections are lost.
It would be better to have both directors working in
normal operation and during failover, to switch the
dead one out. This way, directors are running at
half capacity only during failover.
One way of doing this would be to have two IPs for the
site in the DNS tables, and for each director to serve
up the content through one of the VIPs (the real-servers will
have 2 VIPs). During failover, the up director assumes
the extra VIP and only half the connections are lost.
Presumably if you use 5 directors, then only 20% of the
connections are lost on failover.
Joe
--
Joseph Mack PhD, Senior Systems Engineer, Lockheed Martin
contractor to the National Environmental Supercomputer Center,
mailto:mack.joseph@epa.gov ph# 919-541-0007, RTP, NC, USA