LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: LVS use for mass-webhosting companys: Does it make sense ?

To: "Joseph Mack" <mack@xxxxxxxxxxx>
Subject: Re: LVS use for mass-webhosting companys: Does it make sense ?
Cc: <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
From: "Jochen Tuchbreiter" <jochen@xxxxxxxxxxxxxx>
Date: Fri, 14 Apr 2000 13:56:01 +0200
Hi,

thanks for you comments/suggestions. I do have some questions left though:

[del]
> > Do you think that such a setup would prove more stable than the
> > "standalone"-Server solution ?
>
> If by stability you mean % uptime as seen by the user, yes. However as

the things I worry about most:

- if one machine gets cracked, the cracker is able to deface all sites we
host since every realserver has access to the whole nfs-volume

- if one user is able to write a cgi that takes down a whole box then he
will be able to take down all boxes in the cluster one after another - this
may reduce uptime as compared to a "many isolated boxes" solution


> with all cluster solutions, the complexity is high, you have to setup and
> check that machines can failout safely before you put the LVS online. This

Well, I think that I am brave enough to try this :) We will have to do a lot
of testing before going into production, no matter if we do clustering or
not.


> > Do you know a way to keep two NFS-Servers in sync in realtime ? I am
> > insecure about using coda, since it is still considered "not ready for
> > production use".
>
> there is info on setting up a coda LVS on the LVS website. True it's
> experimental.
>
> one of the LVS developers did a demo with gfs to keep files in sync
> see http://ultramonkey.sourceforge.net/
>

> LVS requires that all realservers see the same file content
> and you want to do this by nfs introducing the fileserver as a spof.

That´s exactly what I want to avoid by having two NFS boxes which are in
sync almost in realtime. The webslaves will check if the nfs-server does
respond and will mount the "spare" server in case of a failure.

I wonder if File I/O will be a weak spot on such a system - this will
certainly be a limiting factor for the cluster-size.


> I don't know how much disk space you're offering, but can you
> setup the same files locally on each realserver. If the files are

Unfortunately this will not be possible since changes on files have to be
visible from all realclients immidiately. Unfortunately I can´t use any
software considered "unstable" like Coda. GFS does not seem to be ready for
production use too. I guess I´ll have to look for some other solution
probably at the expense that both fileserver will not sync in realtime but
every hour or so.


Greetings,

Jochen




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