[Vserver] Hosting Vserver's on CODA or Lustre Distributed File Systems?

From: EKC <webmaster_at_generalsynthesis.com>
Date: Mon 14 Nov 2005 - 16:42:22 GMT
Message-ID: <355a4e960511140842w27914adfj30bc3ff690dda6e3@mail.gmail.com>

Hello,

I have a group of nine Linux boxes each running several Vservers. Each
Vserver is a self-contained webhost with Apache, MySQL, Qmail, etc...
The Vserver "images" on each machine reside on the local hard disk.
There are no network shares (AFS, etc...)

Needless to say, it is becoming increasingly difficult to backup the
Vservers. I am also concerned about redundancy and failover. Having
the Vserver images all spread out is also an administrative pain. I am
not using RAID and would like to use some sort of distributed
data-store for the Vserver "images".

I have looked at Redhat GFS, but ruled it out because of the high-cost
of using a SAN and scalability problems. I would like to scale to
several dozen machines running Vservers.

The ideal solution would be a highly-scalable distributed file-system
with built-in data redundancy (RAID-1 striping across the storage
nodes). I would like to be able to shutdown a Vserver on one machine,
and start it up on another without having to SCP the vserver directory
over.

So far, I have come up with only two solutions:

The first is using CODA. I would set aside a group of 3-4 servers
(including replicas) to host the CODA file-system. The Vserver images
would be stored here. The "client" machines would mount the CODA
file-system and run the Vservers. CODA includes a persistent local
cache on each client to deal with latency issues. However, I'm worried
that CODA may not scale well (based on discussions on the CODA mailing
list).

I am also looking at using the Lustre distributed file system
(http://www.lustre.org). Lustre seems to be actively supported, and is
highly scalable (1000's of nodes). However, it does not yet support
RAID-1 striping across storage nodes or persistent local cache's on
the clients. So, I'm worried that latency may be an issue. Throughput,
however, scales linearly on a Lustre cluster with the number of
storage nodes (read: machines) -- a characteristic that CODA does not
have. If Lustre is the best option, I could always go for software
RAID on each storage node until Lustre supports RAID-1 striping.

Has anyone run Vservers from a CODA or Lustre file system? If so, what
is your experience? Does it scale well? Does CODA have scalability
problems? Does Lustre have latency problems?

Should I use CODA or Lustre?

All advice/input is highly appreciated!

Thanks in advance
_______________________________________________
Vserver mailing list
Vserver@list.linux-vserver.org
http://list.linux-vserver.org/mailman/listinfo/vserver
Received on Mon Nov 14 16:43:51 2005

[Next/Previous Months] [Main vserver Project Homepage] [Howto Subscribe/Unsubscribe] [Paul Sladen's vserver stuff]
Generated on Mon 14 Nov 2005 - 16:43:57 GMT by hypermail 2.1.8