[OpenAFS] Two questions AFS as a *large* webserver system backend in lieu of NFS

[Michael Loftis]

I'm working on the details needed to deploy AFS for my company as a backend 
for our webserver system.  Currently we use NFS and bind mounts to produce 
user 'jails'.  this works fine except every webserver has to know when user 
is created or deleted in order to bind_mount /mnt/jail into ~user/.renv. 
This seems like an idea solution for AFS, just use fs mkmount to centrally 
administer who has .renv's and who doesn't.

What I want to know is how well this may work for us?  Currently I'm 
looking at 4k+ domains, and around 3k actual users with about 90Gb or so of 
actual storage on one NFS server.  We want to split this into two AFS 
servers (build, deploy first afs file, database server, then do second 
file/database server and third database server, followed sometime later by 
other fileserver(s)).

Is AFS going to handle this sort of access pattern fairly well?  We plan on 
using the 40Gb HDDs on eachof the webservers as mostly cache, say 20Gb or 
30Gb.  This will (theoretically) reduce the load on the fileservers and 
internal network.

With the was this is going to work out the system will potentially have a 
500k or more volumes in the cell (minimum one per user, but most users will 
have atleast two or more, one for the user, one for their config area, one 
for each of their websites, maybe one for their tmp filesystem) probably 
using 'balancer' to help manage balancing of usage across fileservers and 
partitions.

Comments welcome, privately though unless the list is interested.

TIA!

Michael Loftis