[prev in list] [next in list] [prev in thread] [next in thread] 

List:       hpux-admin
Subject:    [HPADM] SUMMARY Printing performance
From:       tensor () tensorex ! com
Date:       1997-08-29 1:02:44
[Download RAW message or body]

Sorry for the extremely late reply but had related problem with Glance

not running so there was no real way to measure performance improvement

after trying out the suggestions offered by the group. Still have problem

with Glance but that's another question. Anyways, with some "REAL"

performance numbers (ie loudness of my users moaning!! :) ), here are the 

summary of the suggestions :


Original question was printing of large batch jobs causes users on X-terms

to experience jumpy mouse behavior


Thanks to :	

		Marty Freitas

		Eef Hartman

		Julius Szelagiewicz


[MARTY]

I'm not expert here, but I wonder if it might help to renice lpsched

so it's at a higher priority, allowing the other stuff to run first.


Also, depending on how tight you are for memory, and how big your print 

jobs are, they might be forcing user processes to swap out from memory

to swap space.


[EEF]

> We do suffer a system slowdown due to NFS transfers to the print 

> server whenever we batch large jobs, however according to GLANCE

> the network is doing great with no collisions.  What I mean by system


NFS is well known to have a bad perfomance on WRITING to a remote disk.

Reading is almost as fast as reading a local disk, but writing is much

slower as a lot of status checking, locking etc has to go on on the remote

server.


and many thanks to [JULIUS] for the best solution to my problems


since all is switched, than the only (and best) way to speed up the print 

is to create print files on the SAME file system the spooler is using, 

and that is, come to think about it /var. the spooler creates files in 

/var/spool/lp. now, links are created instead of copying files ONLY if 

the same file system is used for your print files and for temp files that 

spooler works with (the temp files are needed, cause you can delete a 

file after you schedule it for printing and it still needs to be printed) 

in general you want to minimize the amount of copying (i got 20% better 

response time just for forcing the creation of print files in the /var. 

you also want to minimize the amount of network traffic. segregating 

printers onto their own segment is probably the best bet. good luck, julius



[SOLUTION]

My best solution was since that I had more than enough memory and my network

was fine (ie switched and segmented correctly) it was more of an NFS write issue and \
by creating print files on the same system as the spooler for me

on HPUX 9.05 /usr/spool/lp links are used instead of true copies so NFS

traffic is lessened, my users are not feeling so "jumpy" !! It really works !!

Anyone with printing problems should try this suggestion out.


now on to my glance problem ...

******************************************

*	<bold>Tensor Engineering</bold>			*

*	2060 South Patrick Drive		*

*	Indian Harbour Beach, FL 32937	*

*	Phone	: (407) 773-3036		*

*	Fax	: (407) 773-3038		*

*	Email	: tensor@tensorex.com	*

******************************************

--
             ---> Please post QUESTIONS and SUMMARIES only!! <---
        To subscribe/unsubscribe to this list, contact majordomo@cv.ruu.nl
  Name: hpux-admin@cv.ruu.nl                 Owner: owner-hpux-admin@cv.ruu.nl
 
 Archive:     ftp.cv.ruu.nl/pub/digests/hpux-admin       (FTP, browse only):
   *NEW*:     http://www.dutchworks.nl/htbin/hpsysadmin   (Web, browse & search)


[prev in list] [next in list] [prev in thread] [next in thread]