|
Joseph Mack PhD, High Performance Computing & Scientific Visualisation
LMIT, Supporting the EPA Research Triangle Park, NC 919-541-0007 Federal
Infrastructure Contact-Ravi Nair 919-541-5467 - nair.ravi@xxxxxxx,
Federal Visualization Contact - Joe Retzer, Ph.D. 919-541-4190 -
retzer.joseph@xxxxxxx
lvs-users-bounces@xxxxxxxxxxxxxxxxxxxxxx wrote on 05/13/2005 10:39:54
AM:
> I'll re-read the document but I'm sure it implied a read
> only mount with sharred storage would be OK.
even with read-only, the problem is that any particular file
(eg foo.c) will have a different file handle on each realserver
and when you fail-out the realserver, the client will ask the
new realserver (via the director) for the old filehandle.
I've talked to several authors of NFS code (at OLS and via
e-mail) and there's no intention of changing this property.
It seems to be regarded as a "Good Thing". For reasons I
don't understand, the generation of the file handle is buried
so far down in the code that you can't change it without
ripping out lots of other code. It's not obvious to me
why you just can't substitute a new piece of code for
the procedure that generates the filehandle and why it can't
generate a unique filehandle from the /directory/filename.
These problems presumably are why NetApp and Auspex fileservers
are so expensive - they had to write their own version of nfs.
Auspex spent millions$ trying to write ServerGuard and failed.
(I was working for Auspex at the time - the old owners who
started the company retired and the company was taken over
by people who AFAIK just let the company run down and go bankrupt -
morale was low, people were leaving left and right, bad technical
decisions were made everywhere and it's possible that ServerGuard
would have made it just fine under the old owners).
Joe
|
