On Mon, 28 Jan 2008, isplist@xxxxxxxxxxxx wrote:
Depends. For instance, there are many people that use NFS as a backend
for a cluster of mailservers or even webservers.
I've read so much about NFS not being a good choice for production services
yet I know it can be very reliable as I use it around the network a lot. Since
I already have fibre though, that is the way better solution for my cluster at
least.
Whether NFS is suitable for your production environment is 100% dependant
on what you are trying to do. If you have a LAMP cluster/grid that doesn't
require reliable file locking, you may actually find that NFS ends up
being preferable to something like GFS. The major failing of NFS is the
ability to have reliable file locking. This is largely an issue for (but
not limited to) things like DBs. If you don't need this, then it'll
probably be OK.
have IBM's which have 8-way CPU's and can have up to 64GB of
memory.
wow, christ. I don't think you need that much woom for I/O :)
Don't know yet, that's why I asked :). I've got various types of hardware I
can use. Off the top, without having played with this yet, I don't know how
much I/O there might be. Could a single aggregator handle 50 servers looking
to it for their data? I don't know, sure doesn't sound like it could to me.
How I/O bound do you expect your front ends to be? If each of your LAMP
front ends can handle 100Mb, then it'll take 10 of those to saturate a
single 1Gb interface. If your aggregator has a 10Gb interface on each
side, that's 100 servers before you run out of I/O, assuming they are all
going flat out (which let's face it, they won't be). And when you run out
of that, you start adding additional 10Gb NICs to suit.
On the other hand, I've not looked around yet to see if there is in fact
hardware being manufactured specifically to be used in this manner. Or, I can
build something custom based on the requirements.
That's pretty much it. All of this is about primitives / building blocks
being available to put together whatever sort of a solution you deem most
suitable for your particular application. :-)
Right now, aside from building that machine, I need to better understand if I
need a single volume which can grow or a virtual RAID array. Seems a volume
would do the job since the machines are already RAID.
Machines may have RAID on them, but if you lose a whole machine, what
happens then?
Gordan
--
Linux-cluster mailing list
Linux-cluster@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/linux-cluster
