[OpenAFS] newbie questions

Lars Schimmer l.schimmer@cgv.tugraz.at
Tue, 11 Nov 2008 13:13:31 +0100

Hash: SHA1

Papp Tam=E1s wrote:
> hi all,
> I'm sorry if this email contains too simple questions and problems. I'm
> a really newbie to openafs and it was a hard work to be here with afs
> setup, it's a bit strange to me.
> I've made a test setup using this howto:
> https://cs.stanford.edu/wiki/csdcf/Projects/KerberosAFS (+ howto from
> openafs.org, and many many fight with the setup).
> The system is based CentOS 5.2 + official 1.4.7 rpms.
> - there is admin user
> - there is test user
> - there is one partition, which is the root.cell and root.afs (do I say
> well, are them same?)

No, one is /afs and one is the "root" of your cell, in my case root.afs
is /afs and root.cell is /afs/cgv.tugraz.at.

> - I can connect with windows client
> Here are the problems:
> 1. windows client is mapping a drive, I can write is, but it's so slow,
> it's started with ~4 MByte/s and ended with ~11Mbyte/s (~100 Mbit/s). I=
> there any limit or something like that?


If possible, try to disable the crypto option for the client.
Without crypto I reach 30-50 MB/sec on gbit network (which is limited
mostly by the client hardware).

> 2. How can I add a new partition to the cluster for extending the
> storage size? I gues it should be mounted on /vicepa, it's done.

Each fileserver could host up to 256(?) partitions /vicep(a-zz).
Just add a new partition /vicepb and add it to the filserver and restart
OpenAFS fileserver.
The best solution to maintain load balancing and failure security is to
add another fileserver (or even more), each with its own /vicepa.

> 3. What happens, if the somehow one of partitions disappears (eg. hdd
> get wrong, like a node server),l what is the right protocoll in this
> case, is there more possibilities? Did I miss something is the docs?

If one partition went away, it is no more reachable. But if you setup
another FS or another partitions (and created ReadOnly copies of your
ReadWrtite volumes on that partitions) OpenAFS will use the still
available RO copies on the remaining partitions.

> 4. Is there any piossbility to replicate meta data's (root.afs?), and
> also is there any failover solution about it? I'm sorry, I'm working
> only with Lustre, I know only its terminology.

Replicate ONLY MetaData is not possible in usual OpenAFS (imho it is
with the Object Storage system).
The usual way in OpenAFS is to create volumes and mount the volumes into
the AFS space (in my case e.g. volume "homes" is mounted on
/afs/cgv.tugraz.at/home). The volume itself contains the data,
directories and the quota and ACL information.
The volume resist on one of the partitions your fileservers are providing.
OpenAFS provides functions to replicate those volumes and copy the
replicated volumes onto the same and/or other partitions.

Which ends up in: you need to to some magic by yourself, but if done
right, one fileserver can go down and users does not notice at all.
And you can do load balancing fairly easy (even via worldwide slow lines).

> Thank you,
> tamas

Lars Schimmer
- --
- -------------------------------------------------------------
TU Graz, Institut f=FCr ComputerGraphik & WissensVisualisierung
Tel: +43 316 873-5405       E-Mail: l.schimmer@cgv.tugraz.at
Fax: +43 316 873-5402       PGP-Key-ID: 0x4A9B1723
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org