Recommendations for a large implementation

To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Subject: Recommendations for a large implementation
From: Clint Byrum <cbyrum@xxxxxxxxxxx>
Date: Tue, 21 Feb 2006 15:44:43 -0800
Hello LVS users and developers. I have a situation where I'm setting up
LVS for a short term need, so let me explain.

We are an ASP that provides services to newspapers and the like. To
date, we've been using highly redundant apache servers to load balance
requests using mod_backhand, and no LVS-like system. However, HTTP load
has started to get high, and we'd like to distribute things out at a
lower level.

We're serving around 15-20MBits of sustained HTTP traffic, right now
through several redundant proxies/app servers, with a backend of 15-20
search machines. Traffic is partitioned logically among these machines,
so load isn't perfectly balanced. We're going to flip this a bit, and
make the first request happen at all machines, and proxy the read-write
requests down to other application servers when necessary.

We're talking with F5 about doing a BigIP implementation, but we are on
a time crunch, and know that LVS will probably get us where we want to
be in the short term. So, without further ado, the questions:

- We can make all of the HTTP decisions we want at the server level
using mod_proxy. However, ktcpvs can also help with these, right? Is it
stable? And how much CPU overhead does it incur at the load balancer
level? Annecdotal evidence would be helpful here.
- We'll be using ldirectord and heartbeat to manage failover. Is that
still the best way to go?
- What kind of hardware requirements would you, the community suggest
for this solution? We were looking at using HP DL-360 G1's with a single
- Ok this is highly subjective, I know.. but how does LVS fare against
the BigIP-1500. Anybody replaced one with the other?

Any guidance is much appreciated!

<Prev in Thread] Current Thread [Next in Thread>