LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: scheduling not based on connection

To: Vladimir Dzhuvinov <vd@xxxxxxxxx>
Subject: Re: scheduling not based on connection
Cc: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
From: Joseph Mack <mack@xxxxxxxxxxx>
Date: Wed, 15 Nov 2000 07:20:26 -0500 (EST)
On Wed, 15 Nov 2000, Vladimir Dzhuvinov wrote:

> 
> 
> Hi,
> 
> I'm relatively new to LVS, been through a number of docs and examined a
> small Ultra Monkey LVS in action. Good first impressions - thanks to
> the LVS authors :-)
> 
> I was considering a case where LVS could be successfully used, however
> it appears to me that the scheduling algorithms are connection oriented
> and as such most suitable for network server apps. My case is the
> following: 
> 
>       * 5 compute servers 
>       * each server is used by a group of people to run 
>         various computing jobs via telnet, files mounted via NFS
>       * each group server has a highly irregular load pattern, 
>         depending on the group's project
> 
> Obviously, it would be a good idea to introduce a load balancer to
> spread the load between the five servers. 

I assume each application runs on only one node. In that case you should
use mosix rather than LVS. With mosix, the users logon to a master node
where the jobs are started. Mosix reassigns the job to the least busy node
in the compute farm. To the users the master node appears to have n times
the compute capacity.

I have not used mosix, but the beowulfers who have used it, like it, i.e.
there aren't stability problems etc.

Joe

--
Joseph Mack mack@xxxxxxxxxxx



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