On Tue, 11 May 2004, Edward Chee wrote:
> I'm trying to find an open source linux based load balancer that supports
> persistent connections & pipelining to mutiple real servers. By maintaining
> persistent connections and pipelining client requests, we minimize TCP
> processing overhead & free up CPU cycles/memory on the real servers.
Squid supports persistent connections, and multiplexing of client requests
over the same server connections, but not pipelining.
The HTTP RFC strongly discourages HTTP proxies from pipelining client
requests due to the nature of HTTP.
> Riverstone networks and Netscaler make boxes that do these but from google
> searches, it seems like they go for somewhere betw 30K-90K! LVS looks like a
> well adopted load balancer with failover - But I can't find any info
> indicating it supports persistent pipelined connections to real servers.
LVS load balances TCP connections. As long as the client (and server)
keeps the connection open it will be there.. but this also means that your
web servers will have as many connections as there is client connections..
LVS does not help you aggregating many client persistent connections on
fewer server persistent connections, but Squid does.
> From scanning the Squid website and archives it looks like Squid might be
> able to do this. I see that Squid can be configured in reverse-proxy mode as
> a load balancer. I've also seen indications that HTTP1.1 persistent
> connections is supported - but I haven't been successful at finding clear &
> definitive documentation on this.
Squid supports HTTP/1.0 persistent connections. Squid does not yet fully
support HTTP/1.1.
> If anyone has done this, I would be most grateful to hear about it.
We build a quite affordable reverse proxy using Squid as the HTTP engine
and load balancer (with https support). Works really well.
http://www.marasystems.com/pdf/e_mara.pdf
Regards
Henrik
Received on Tue May 11 2004 - 14:58:25 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Jun 01 2004 - 12:00:01 MDT