On Wed, 2003-07-09 at 13:02, Ivan Popov wrote:
> On 9 Jul 2003, Dick Kniep wrote:
Hi Ivan,

> 
> Have you considered local boot, while optionally from network for
> (re)setting up the local system? It is quite common and efficient
> approach, especially when the local part of the system is small.

We have tested remote boot extensively, and it works wonderfully.
Because the kernel is only 1 MB, it starts almost as quickly as from a
local disk. Also, because of the superb cashing that Linux is doing,
only the first time you start an application, it takes a little longer,
but afterwards it all works almost as fast as a local machine and in any
case fast enough for the day to day work the users do. Even a heavy
application as OpenOffice runs smoothly on a Celeron 1Ghz with 256 MB.

Has any of you had any experience with loading Coda at startup? Because
the root has to be mounted, and I wonder if that is at all possible with
Venus? Because if so, we could use Coda to boot from....

> 
> > So I'm drawing the conclusion that hardware RAID solutions should be the
> > way to go? (Remember we are trying to solve Single Point of Failures) We
> 
> What do you mean? RAID is in no way a distributed filesystem, why do
> you mention it in the same contents as NFS and Coda?

We are trying to solve single point of failures because of the fact that
we want to boot from the network.... If you use remote boot, nothing
goes anymore if the server is down, so as we are planning to accommodate
some 140 clients the servers reliability should be incredibly high. So:

We can go for a distributed file system like Coda over several servers.
If one server goes down, the users won't notice... 

OR

we could go for hardware related solutions to avoid the single point of
failure of the server, by multiplying servers, but keeping the storage
in one place (a raidcabinet).  

> 
> > should implement a number of systems with LVS, and connect a RAID
> > cabinet to the different systems. The reliability will be less, but I
> 
> To multiple NFS servers or to all clients?

In that case, we'll use NFS only (so no Coda)

> 
> > think it is probably not feasible to get the Coda/NFS solution working
> > in a reliable way without a lot of programming.
> 
> If you mean Coda over NFS, then just forget it.

I love it, when you get a clear answer, even when it is not really what
I was hoping for ;-)

> 
> Best regards,
> --
> Ivan
>