Re: Clustering
- From: Chris Scheers <chris@xxxxxxxxxxxxxxxxxxx>
- Date: Fri, 16 Dec 2005 17:22:41 -0600
Bill Gunshannon wrote:
In article <FD827B33AB0D9C4E92EACEEFEE2BA2FB773691@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx>, "Main, Kerry" <Kerry.Main@xxxxxx> writes:
As I recall there was a university that ran something like 120 WS's/servers in a cluster.
This brings up an interesting question (at least for me!)
You mentioned WS's above and I assume that means WorkStations. What, if anything, would be the advantage of building a cluster of, let's say, 2 multi-processor Vaxen like I currently have in the department and a dozen or so VS3100's? Could all the VS3100's run diskless getting all their support from the HSJ served disks on the big boxes? Assuming the cluster traffic was all on a private ethernet and access to the world was only through the two big boxes would performance be good enough? Is there something important I missed because I really have no idea how a VMS Cluster works, never having built one but considering it now. (Especially if it can make the whole system more visible locally!)
Then, of course, would come the biggest question. Does HP have a bunch of fully loaded VS3100's with big monitors that they want to truck up here as a donation so I can build a this dream VAX lab. :-)
I set up something like this once. Once of the questions you need to ask your self is: How important is reboot time for the cluster?
I once saw a cluster of 80-90 some VS3100s all being served by a single 8800 as the boot/disk server. From power up to first login on all workstations took over four hours. (A 10MB ethernet can only do so much.)
The cluster I setup up had close to 40 VS3100s. The plan I used was:
1) Every machine has a page/swap disk. (In this case, they were 52MB RZ22s.)
2) Every fifth machine is a boot server. These machines had enough disk space for all software (VMS and layered apps), but no user data. I think we used 200MB RZ24s.
3) Each boot server serves four other machines. All machines in this "sub-cluster" are on the same ethernet segment.
4) The boot servers were kept synchronized with RSM. (Is that product still supported?) Except for SCS IDs and such, all boot servers were identical.
5) Boot servers were used as user workstations, just like the "diskless" machines. In fact, the users were not aware of which machines were boot servers and which were not.
6) All user data was stored on a "disk server" which was not used as a user workstation. This would be your HSJ machines. All shared cluster files (SYSUAF, queue files, etc.) were stored on the disk server.
7) The disk server was the only machine with votes. (If I redid this, I would consider spreading some votes around to help spread lock mastering, but this didn't seem to cause any performance problems for us.) The disk server did nothing but file serving, print queue execution, and lock mastering.
This configuration worked very well. Time from all machines off to login at all machines was about 15 minutes. This was with VS3100-30s.
All machines except the disk server were used as user workstations. User's could use any workstation interchangably.
User machines had no local data, so they did not need to be backed up.
Boot machines were identical to each other, so they could be restored from another boot machine's disk, so they did not need to be backed up.
The only files that needed to be backed up were the user files on the disk server.
-- ----------------------------------------------------------------------- Chris Scheers, Applied Synergy, Inc.
Voice: 817-237-3360 Internet: chris@xxxxxxxxxxxxxxxxxxx Fax: 817-237-3074 .
- References:
- RE: HP announces new Integrity Blade Servers - OpenVMS?
- From: Main, Kerry
- Clustering (was: Re: HP announces new Integrity Blade Servers)
- From: Bill Gunshannon
- RE: HP announces new Integrity Blade Servers - OpenVMS?
- Prev by Date: Re: PHONE error - Invalid specification of node or person. Try again.
- Next by Date: Re: calloc() fails, no more memory
- Previous by thread: Re: Clustering (was: Re: HP announces new Integrity Blade Servers)
- Next by thread: Re: HP announces new Integrity Blade Servers - OpenVMS?
- Index(es):
Relevant Pages
|