Newsgroups: comp.unix.large
Path: utzoo!utgpu!cunews!bnrgate!bwdls61!bwdla31!marmen
From: marmen@bwdla31.bnr.ca (Rob Marmen 1532773)
Subject: Re: nis/yp usage question
Message-ID: <1991Jun24.181644.16815@bwdls61.bnr.ca>
Sender: usenet@bwdls61.bnr.ca (Use Net)
Nntp-Posting-Host: bwdla31
Organization: Bell-Northern Research, Ottawa, Canada
References: <4932@spim.mips.COM> <1991Jun21.140331.14724@bwdls61.bnr.ca> <1991Jun24.030546.19659@lokkur.dexter.mi.us>
Date: Mon, 24 Jun 1991 18:16:44 GMT

In article <1991Jun24.030546.19659@lokkur.dexter.mi.us>, scs@lokkur.dexter.mi.us (Steve Simmons) writes:
> marmen@bwdla31.bnr.ca (Rob Marmen 1532773) writes:
> 
> I think you've got a massive case of overkill here.  My experience was
> that one 3/160 could happily server approx 50 YP clients.  Assuming a
> sun4 is twice as fast (heh), your 2000 workstations should need about
> 20 nis servers.

	There is an overabundance of slaves. However, since I am on 7/24 call,
	I'll take a bit of redundancy in favor of a few more hours of sleep.
	We use a ratio of 1 slave per 25 workstations. This was for the older
	release of HP workstation (most of our slaves are old HPs) and was 
     backed up by test results. The new Sun Sparc 2s can easily handle
     over 80 clients. I am being limited by older hardware and software
     releases (HP NIS code is ANCIENT!).

	Our machines are organized into workgroups that are isolated from
  	the backbone by bridges. To minimize single points of failure, we have 
 	a minimum of 1 nis slave per workgroup ( with an additional slave for
  	every 25 additional clients). A workgroup with 45 clients will
  	have two NIS servers.

	The failure of a workgroup is an automatic severity one problem. You
	don't go home till it's fixed. Therefore, each workgroup needs to be
	an autonomous entity.

	Every couple of months, I re-examine the performance numbers. However,
	I still get performance problems with clients trying to bind to an
 	insufficient number of slaves. I could probably reduce my overhead by 50%
	if I was just running Suns, but I have to work with what I've got.

	I am hoping that HP will finally wakeup and implememt the Sun
	improvements. However, I am not holding my breath on that one!

	So to summarize, I agree with your statement that it appears to be overkill.
	But having gone from total chaos to a stable environment, I'll keep
	the current setup I have. As vendors PROVE that their code can handle our
	applications, then I'll adjust the ratios and distribution accordingly.
> 
> Also, I second the recommendation of the Hal Stern book.  Very nice.

I've had the book on order for several weeks now. Unfortunately, it takes
a while for books to get across the border. It does look like an excellent book.
( Care to ship me a copy? ;-) )

My apologies to all if it appears that I am ducking an issue. However, corporate 
policy ( yes, that ugly demon again) forbids me from disclosing actual numbers
or describing the topology in detail. I like my job and it does pay the rent.

cheers, 

rob...

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
| Robert Marmen             marmen@bnr.ca  OR             |
| Bell Northern Research    marmen%bnr.ca@cunyvm.cuny.edu |
| (613) 763-8244         My opinions are my own, not BNRs |
