|
On Thu, Dec 18, 2003 at 10:42:52AM -0800, Brett E. wrote:
> ntadmin@xxxxxxxxxxxx wrote:
> >
> >Horms,
> >
> > Yes, I have actually just tested the failover with a fully upgraded
> > LVS
> >system using the IPaddr2 script and instead of causing hurtful
> >multiple minutes of downtime it finished the failover in under 3
> >minutes and appears to be functioning as well or better then it did
> >before. Unfortunately we do need to failover this many addresses as
> >we are an ISP and we host 1000+ web sites and several thousand email
> >accounts for hundreds of domains. We use a mix of Micro$oft and
> >Linux platforms and their various technologies. It's a complicated
> >mess and we are still in the process of concatenating services from
> >multiple purchases we have made as a company. When we are all done
> >with this process it is very possible that all of the services will
> >fit within a couple of small ranges but for now.. 'tis the only way..
> >;)
> >
> >Once again I would like to thank everybody who responded to my posts,
> >you were all a tremendous help!
> >
> >Thanks,
> >Billy Olson
> >
>
> FYI, with keepalived, failover takes around 1 second for me on a
> production system with about 2,000 connections/second. Of course I have
> only one virtual IP. I believe keepalived uses gratuitous arp and
> adds/removes ip's via C code using the "ip" method otherwise I'm not
> sure how it accomplishes this feat. It may be worth giving it a try.
Heartbeat could be configured to give the same failover time
for such a situation. The thing that is taking the time is
the large number of addresses that are being failed over.
I suspect keepalived would also slow down, though without
testing it is hard to say how much.
N.B: I don't have anything against keepalived. Actually, I think it is
quite good. But I thought I should point out that heartbeat's
failover isn't actually slow as such. Though it can be, depending
on your configuration.
--
Horms
|
