On Tue, 2 Sep 2003, Cyril Bouthors wrote:
> Ivan Popov writes:
> > you may need about 10 servers (or server processes) to serve 200G
>
> Tu sum up, if I have 2 computers with 100GB each, I need 2x5 ext2
> partitions and 2x5 coda processes, is that right?
It should work. Sorry I did not have time to look at your previous letter
and possibly guess what is an optimal configuration for you.
One server process / served partition per 20Gb, as you write, should work
fine. The main drawback of such configuration is that you'd have to make a
decision at each volume creation - which server(s) shall serve the
new volume, probably depending on free space available.
About your biggest files - a 100Mb file at first access on a client will
probably take over half a minute to fetch into the cache, on a LAN, the
same for first open() if the file has been modified by another client.
So the usage pattern becomes very important.
Repeating accesses from the same client to a "readonly" file will be about
as fast as the local filesystem containing the cache (say ext2).
> If
/home is a symbolic link to /coda, will all my user
directories be > available as /home/$user or as /home/$partition/$user ?
Three completely different things,
- I would advice to use the new realm/cell-aware paths,
so let us think /home -> /coda/<your.cell>/home
- the partition names are not seen in any way on the coda filetree,
not at all.
- you choose a partition and a name for a volume when you create a
volume
- you choose a mountpoint for a (named) volume when you create the
mountpoint
neither partition names, nor volume names are visible in the pathnames
(though it is a good idea to keep volume names in sync with their
mountpoints)
- if at all possible, do not use "short names" for files on Coda,
rather give the users $HOME==/coda/your.cell/home/username
instead of /home/username
It will give you the freedom to avoid (re)configuring all clients
according to your (possibly changing) design choices
Regards,
--
Ivan
