LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: Overload flag is not resetting

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: Overload flag is not resetting
From: Joseph Mack <mack.joseph@xxxxxxx>
Date: Wed, 02 Feb 2005 00:09:21 -0500
Kees Hoekzema wrote:
> 

> I check my realservers for their load and adjust the weight of the servers
> around every 10 seconds.

Hi Kees
        It seems you've found a bug here. Hopefully someone will look into it.
        I'm one of the people who think you shouldn't dynamically change the 
weights of
your realservers unless you've got a real good reason. I know Jeremy Kerr did 
an honours
project on this topic, so he's probably looked into the control and feedback 
theory on the matter
and knows more about it than I do.
        By its design LVS loadbalances. You should assign weights according to 
the speed
the realservers handle the requests. If all realserver have the same hardware, 
then their
weights should be the same. You should only change change the weights if 
something about
the traffic you're handling breaks the assumptions of LVS. These assumptions are

o clients come from random IPs

o clients make similar requests.

The first problem is the AOL proxy problem. The first machine to get a request 
from an AOL client
will service all the AOL clients.

The second problem would be if the LVS is handling a mixture of short and long 
requests (eg
http hits and long lasting tcp connections like telnet or https). All the http 
hits will be 
serviced in the same time, but the https and telnet sessions will vary in 
connection time by
a large factor. If the variation causes problems, then I imagine you could use 
dynamic weighting.
Still it might be simpler to add an extra realserver or two to handle any extra 
load.

If you're going to dynamically reweight your machines, then you should do it on 
a timescale that
is long compared to the events that they are handling, ie if you're handling 
http hits, then the
calculation of the new weight should sample the load no more than every few 
secs. If the load comes 
from https, then every few minutes at the most.

Ratz says that his reweighting has time constants of days.

No-one (except perhaps Jeremy) has done a study of the benefits of dynamic 
weighting,
so my statements are just theory at the moment.

Joe

-- 
Joseph Mack PhD, High Performance Computing & Scientific Visualization
LMIT, Supporting the EPA Research Triangle Park, NC 919-541-0007
Federal Contact - John B. Smith 919-541-1087 - smith.johnb@xxxxxxx

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