[SUMMARY-2] High Availability Cluster for NFS

Reasons for my updated summary:
* I got few more responses. 
* Attachments sent had redudant original message. 
  So I cleaned things up now.

Three of recent responses mentioned that response showed in item 5
in my original SUMMARY is incorrect. Reasons:
i) NFS is stateless protocol
ii) Sun cluster works by moving the server (the logical host) from 
one system to another, taking the service (HANFS) with it. The 
clients hardly notice any change.

Additional High Availability Cluster software:
RSF-1 from http://www.high-availability.com
Runs on Solaris. Priced, I think; waiting to hear back.

Original SUMMARY:
-----------------------
Thanks to:
Harrington, David B[David.Harrington@dscr.dla.mil]
Timon, John [John.Timon@labatt.com]
Mohan Doraiswamy [mohan.doraiswamy@overture.com]
Debbie Tropiano [debbiet@arlut.utexas.edu]
Liz Maidment [liz@stealth-cow.freeserve.co.uk]
Yura Pismerov [ypismerov@tucows.com]
ilya.birman@abnamro.com
John Martinez [john@mtbiker.net]
Bertrand_Hutin@notes.amdahl.com

Brief:
1. Veritas Server Quick start
   Veritas Cluster Server software (VCS) 
   Veritas SANPoint Foundation Suite HA (using
   Cluster File System and Cluster Volume Manager)
    - priced.

2. Sun Cluster Software 
    - priced and is available cheaper via ScholarPac for
      educational institutions. I am checking into this.

3. Legato Cluster products

4. Through action scripts 

5. One response doubts that NFS will not work. He says:
"The thing is, when NFS server goes down client needs 
to re-mount the partitions"


Detailed:
I have attached responses that were some what detailed than above.

Sundar

Original message:
---------------------------
I am looking for 3rd party software (freeware, if possible) solution under
SOLARIS. Here is our problem.

We would like to have a system with no single point failure primarily for
NFS services. We are thinking about two Dual CPU 64-bit Ultrasparc processor
systems with heartbeat approach and common storage. Solutions from Sun
Microsystems are way too expensive for us. I would like to know if there are
any 3 party companies offering a "Software" on SOLARIS to check heartbeat of
machines and to do the switch over automatically if one of the two systems
should fail. Any web links will be helpful, especially if there are any
freeware packages.

I can find many useful sites for LINUX. Unfortunately, I need ones for
SOLARIS. 

Any experience on the above subject is welcome. Thanks in advance.


-----------------------------------------------------------

From: Debbie Tropiano <debbiet@arlut.utexas.edu>

Have you checked Sun ScholarPac pricing for Sun cluster software?
Apparently it's now available and much more reasonably priced
(but only if you're doing it under ScholarPac -- and I would think
that iupui.edu would already have a ScholarPac program or would
qualify for one).

Alternately, the Sun Fire V880 systems have a great deal of redundancy
built in (plus you can impletement IP multipathing and disk multipathing
for not too much extra).  While there are still a few single points of
failure, it greatly minimizes it while greatly minimizing the effort
needed to manage the systems.

Just some ideas based on our experience in trying to solve what appear
to be similar problems.

---------------------------------------
 
From: John Martinez <john@mtbiker.net>

I implemented a four-node NFS cluster on Suns using Veritas SANPoint 
Foundation Suite HA.

It is not cheap, but it works well and is rock solid.

I am employing Clustered Filesystem and Clustered Volume Manager (CFS 
and CVM), which allow you to mount the filesystems on all nodes in the 
cluster at the same time from the SAN storage system. This is very 
useful to reduce fail-over time.

I can let you know all about it, if you want and give you an unofficial 
reference.

Let me know and I'll give you my contact info.

--------------------------------------------
From: "Timon, John" <John.Timon@labatt.com>

gut instinct, buy Sun Cluster or Veritas Cluster

now that I have said that, budget concerns are very real so I will continue
under that premise.

Cluster concepts

- multiple hosts, checking to make sure each other is alive.
- share disk of some sort
- one host goes bye-bye, the other assumes its job (virtual IP address moves
over)

there are some very serious caveats if you want to grow your own cluster.
First and fore-most is reliability.  

if you want to do this for an NFS server, you will need both machines able
to talk to the same disks. (this too costs money)

when one machine goes down, the other will kick off a series of scripts to
startup a virtual IP and then bind the NFS server to that virtual IP.
routers/switches will need to have their arp caches flushed, since network
traffic is done to the MAC, not the IP, and arp will associate MAC to IP.

in theory it could be done by scripts in the solaris world.

I think that unless you have decided that Linux will not enter your shop,
you may want to investigate the feasibility of Linux clustering (I am not
very familiar with this) it will likely be cheaper.

But in any event there will have to be some sort of outlay of Cash (either
in shared disk {SAN, T3, A5x00, etc}, or proper clustering software)

----------------------------------
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers