|
You may want to consider a combination of cluster technologies to
accommodate your needs. For example, combine an LVS cluster with a high
availability option such as MC/ServiceGuard from Hewlett Packard
(http://www.hp.com) which is designed to provide fault tolerant access
by sharing a common disk system across multiple nodes in a cluster.
Mission Critical Linux have a similar product known as Convolo, as well
as an open source version called Kimberlite (http://www.mclx.com/ for
details).
In it's most basic form, you have two nodes which share an external disk
system via SCSI or Fibre Channel. One node exports half of the shared
disk, the second node exports the other half. Each node monitors the
other, and if node A dies, node B takes over all the tasks of node A as
well as continuing to provide its own. We use this topology here to
support our manufacturing data servers, and it is very successful.
Storage is only the first part of the equation, though. If you want to
load balance WWW access, then LVS is the tool for the job, no question.
The idea might be to use an LVS cluster as the interface to the storage,
creating a front-back model for your application. Clients access the
data by connecting to the LVS virtual IP, with each node having equal
access to your data storage cluster.
Another option would be to use a shared storage pool file system on an
LVS cluster, by deploying GFS (Global file system,
http://www.sistina.com/products_gfs.htm), or Intermezzo
(http://inter-mezzo.org/) on each of the real-servers in the cluster.
Best results seem to come from using a fibre channel back-end for these
sorts of file systems, though (especially with GFS). And I'd prefer the
stability of the front-back model, for my preference (the products on
offer for this kind of environment are more mature).
--
Malcolm Cowe.
IT | Technical Computing, Telephone: +44 131 331 6466
Agilent Technologies Ltd. Telnet: 313-3466
"alpha1976@xxxxxxxxx" wrote:
>
> I'm a student of Milan's University and I'd like to ask you some
> information about how to deal with an academic project, which involves
> a lot of problems.
> I have to plan a very big and "important" server which should contain
> and manage multimedia data, such as audio files, video files, images
> and so on.
> This server should work as a "fileserver" for a very important library
> (that will probably be built in a couple of years in Milan) and should
> be accessed from Internet too because users may access video files by
> streaming technology. The documents stored there, will be "linked" to a
> big and complex database system, which is contained in another server.
>
> I know that is required a lot of disk space (because of the kind of
> data that is stored) and I also know that a load balancing is needed.
> It's also important to include backup sistems (such as tapes) to
> preserve data and support for hard disks mirroring.
>
> My idea is (for the File Server) the following: 1 master station and 5
> cluster nodes with the same hardware.
> Each node has 4 SCSI hard-disks (180 GB on each HD) configured with
> RAID-5 technology (3HD for "real" storage and 1HD for parity storage).
> My question is: can master station (in a LVS) recognize ALL the HD
> placed on the nodes as a unique and big drive (the total size is abuot
> 2.5 TB)?
> I need some special SCSI card or everything is managed by LVS software?
>
> Thanks for your help!
>
> _______________________________________________
> LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
> Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
> or go to http://www.in-addr.de/mailman/listinfo/lvs-users
|
