LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: Fundamental performance limits of a linux based load balancer / serv

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: Fundamental performance limits of a linux based load balancer / server system?
From: Sébastien BONNET <sebastien.bonnet@xxxxxxxxxxx>
Date: Tue, 11 May 2004 11:43:38 +0200
(1) Load balancer bottlenecks

(a) max open connections is limited to the number of available ports (65535-1024 divided by TIME_WAIT). For systems with TIME_WAIT set to recommended 4 minutes, this limits max open connections to measly 268. See http://support.zeus.com/doc/zlb/sizing_guide.pdf

Unless I'm mistaken, there's no "open" connection, just tracked connections. The clients are not "connected" to the load balancer.

And BTW, your computation is only valid for a client, not a server. Assuming a client always uses a different port for an outgoing connection, it can roughly initiate 65K connections.

On the server side, there's no port limit for a daemon listening on a single port: it uses just one ! The port used by a client is bound to the client machine, not the server. Several clients can have the same source port.

If there was no file descriptor limit nor memory constraint, a server could handle way more than the current "port limit" (65K) simultaneous connections.

(2) Server bottlenecks
(a) Linux max open connections due to available ports or file descriptors on the web server.

Right for file descriptors, wrong for open connections.

--
Sébastien BONNET
<Prev in Thread] Current Thread [Next in Thread>