|
If your problem is to gain high availability more than just load balance, and
you are not tight to NFS, I would suggest to turn to Coda (there is a link to
it's home site on linux-ha.org) : this distributed file-system allows to
distribute read-write file access to different hosts, using the network. Each
host must install the Coda client to have access to files, and no LVS will
occur there.
You can read a good presentation of Coda on http://www.mcc.ac.uk/~zlsiial/coda/
Nicolas Huillard
-----Message d'origine-----
De: Paul Lussier [SMTP:pll@xxxxxxxxxxx]
Date: jeudi 11 mai 2000 17:52
À: Jeff Langley
Cc: 'lvs-users@xxxxxxxxxxxxxxxxxxxxxx'
Objet: Re: Possible to Load Balance NFS.
In a message dated: Thu, 11 May 2000 10:30:04 EDT
Jeff Langley said:
>Hi Everyone,
>
>I am researching the possiblity to do load balancing of NFS both dynamic
>and direct type automounts.
The problem with load balancing NFS is that you then need to have more than 1
system accessing the same physical drive (or RAID array), *or* have multiple
copies of the data scattered among different machines.
With the latter scenario, you end up with an updating problem. Typically,
this type of configuration would only be used if the data being served were
essentially static, which implies that it's not to be used for things like
home directories, development areas, mail/print spools, etc.
If you are planning on using it this way, you can do some sort of load
balancing with NFS/Automount natively. For example, and automount map like:
/static/stuff systemA:/stuff,systemB:/otherstuff,systemC:/morestuff
would tell the client to mount /static/stuff from whichever system responded
first, which would presumably be the machine with either the lightest load or
fasted network connection. Granted, it doesn't always work that way, and LVS
may well be a better way to do this.
The former configuration though, where you need constantly changing dynamic
data being shared is quite tricky. 2 systems must not only access the same
drive/array/bus, but must also be simultaneously *changing* that data. This
isn't good, since neither has any idea of what the other is doing. Making the
systems aware of each other, I would guess would be more in the scop of a
kernel development group, not the LVS/HA groups, but I might be wrong, being
new to this group myself.
I would enjoy more discussion on this topic, as I'm not all that knowledgable
on how one *could* load balance NFS among multiple systems. Especially if
there are ways around what I've already mentioned.
Thanks,
--
Seeya,
Paul
----
"I always explain our company via interpretive dance.
I meet lots of interesting people that way."
Niall Kavanagh, 10 April, 2000
If you're not having fun, you're not doing it right!
|
