roy@alanine.phri.nyu.edu (Roy Smith) (07/19/90)
I'm installing C-news for the first time on a Sun file server. I havn't made up my mind yet whether I want to have all the diskless clients NFS mount the news spool directory, or if I want them to talk NNTP to the server. I already have rrn running on the clients, using a different server, and I'm going to have to support NNTP on this server anyway (for news transfer), so it's not a problem of getting the software up. Given a free choice between the two methods, which would you pick? Is either particularly more efficient in terms of CPU time or network bandwidth? -- Roy Smith, Public Health Research Institute 455 First Avenue, New York, NY 10016 roy@alanine.phri.nyu.edu -OR- {att,cmcl2,rutgers,hombre}!phri!roy "Arcane? Did you say arcane? It wouldn't be Unix if it wasn't arcane!"
emv@math.lsa.umich.edu (Edward Vielmetti) (07/19/90)
In article <1990Jul18.233115.25423@phri.nyu.edu> roy@alanine.phri.nyu.edu (Roy Smith) writes:
I'm installing C-news for the first time on a Sun file server. I
havn't made up my mind yet whether I want to have all the diskless clients
NFS mount the news spool directory, or if I want them to talk NNTP to the
server.
I'd do strictly NNTP, that's one less mount point on all your clients. If
at some point you want to run all your news on a 3/50 in a closet, you don't
have to go around changing everything either. Systems which would not otherwise
have to mount that file server / news server won't hang with NFS timeouts if
it goes down, they'll just get NNTP server unavailable.
Only disadvantage I can see is that 'rn' is somewhat less nice to use as an
NNTP client cause it doesn't automatically notice new news; depends on what
people are used to.
--Ed
Edward Vielmetti, U of Michigan math dept <emv@math.lsa.umich.edu>
comp.archives moderator
jim@cs.strath.ac.uk (Jim Reid) (07/19/90)
In article <1990Jul18.233115.25423@phri.nyu.edu> roy@alanine.phri.nyu.edu (Roy Smith) writes:
I'm installing C-news for the first time on a Sun file server. I
havn't made up my mind yet whether I want to have all the diskless clients
NFS mount the news spool directory, or if I want them to talk NNTP to the
server. I already have rrn running on the clients, using a different
server, and I'm going to have to support NNTP on this server anyway (for
news transfer), so it's not a problem of getting the software up. Given a
free choice between the two methods, which would you pick? Is either
particularly more efficient in terms of CPU time or network bandwidth?
It's probably better to use NNTP than NFS to let clients read news.
News reading causes a lot of recursive directory traversal. It is best
that NFS avoids this sort of thing because it generates large numbers
of NFS requests which will hammer the NFS server... [Compare the load
on the server of running du of /usr/spool/news locally and then trying
it from an NFS client that has mounted this directory....]
Jim
tale@cs.rpi.edu (David C Lawrence) (07/20/90)
It dpeends largely on where you can spare the resources. You can hammer the server a lot more by having upwards of 20 nntpd processes running on it simultaneously than you would by having 6 or 8 nfsds. The trade-off with using NFS is that the response time will be a little slower even if the load on the server is on the light side. The gain of NFS is that you can locally access all of the mounted directory, not just whatever the NNTP server wants to tell you about (which is pretty much just the active file and the articles). I miss that latter feature on turing here as there are several things that I put in rpi.edu:/usenet, from administrative files to news sources to news information, that I would like to have locally available for mailing to people or leisurely perusal but instead I must log into the server, either directly or through ftp, to transfer them to turing before I can redistribute. A minor annoyance at best but I think I will soon get /usenet mounted here too. -- (setq mail '("tale@cs.rpi.edu" "tale@ai.mit.edu" "tale@rpitsmts.bitnet"))
hirai@cs.swarthmore.edu (Eiji Hirai) (07/20/90)
At our site, we mount /var/spool/news via nfs [also running nn 6.4.9, rn patchlevel 47 and cnews 25-May-1990]. However, I'm sort of convinced that I should make my newsreaders read news via nntp now. There are couple of things that make me hesitate though: 1) Is there any overhead in having multiple nntp processes on the server running? We have a lot of news reading folk on our network and I assume (naively without looking at code) that each nntp news reader will have their own nntp process on the server. Our server only has 28M of memory. 2) I noticed a significant drop in response time when I read news via rrn compared to when I read news via nfs mounted /var/spool/news. Is this general or was it just my case? Halp! -- Eiji Hirai @ Mathematics Dept., Swarthmore College, Swarthmore, PA 19081-1397 hirai@cs.swarthmore.edu | hirai@swarthmr.bitnet | uunet!hirai%cs.swarthmore.edu Copyright 1990 by Eiji Hirai. All Rights Reserved. Permission to reproduce or quote explicitly denied except on Usenet. I don't speak for Swarthmore College.
amanda@mermaid.intercon.com (Amanda Walker) (07/20/90)
In article <JIM.90Jul19175344@baird.cs.strath.ac.uk>, jim@cs.strath.ac.uk (Jim Reid) writes: > It's probably better to use NNTP than NFS to let clients read news. It all depends. In particular, for a server serving many simultaneous clients, NFS is probably better, since you won't clog the server with copies of nntpd. This will reduce the load on the server, and keep performance up. For a big site, I'd say "read with NFS, post with NNTP." -- Amanda Walker <amanda@intercon.com> InterCon Systems Corporation
brtmac@maverick.ksu.ksu.edu (Brett McCoy) (07/20/90)
In <VZVM0L@cs.swarthmore.edu> hirai@cs.swarthmore.edu (Eiji Hirai) writes: >At our site, we mount /var/spool/news via nfs [also running nn 6.4.9, rn >patchlevel 47 and cnews 25-May-1990]. However, I'm sort of convinced that I >should make my newsreaders read news via nntp now. >There are couple of things that make me hesitate though: >1) Is there any overhead in having multiple nntp processes on the server >running? We have a lot of news reading folk on our network and I assume >(naively without looking at code) that each nntp news reader will have their >own nntp process on the server. Our server only has 28M of memory. At our site we have a SPARCStation 1 as our news server. I has 16 meg of memory and also acts as the system fileserver for 4 other SS-1's. The SS-1 is not short on horsepower, but it does tend to run short on memory. With nfs I have 4 nfs daemon's running all the time and handling all of the load. With nntp there is a separate nntp daemon for each person reading news. This causes more memory to be consumed which causes more paging and a general decrease in system performance. I very rarely see the nfs daemons using a noticible chunk of the processor, only when people are logging on to the diskless clients or are editing big files and running large programs. I do notice nntp daemons running high up on the top list and chewing up a sizable chunk of memory. Since news reading tends to be get an article, pause while reading, get another, pause, etc., it stands to reason that a few nfs daemons can service a lot of people more efficiently then a bunch of nntp daemons that try to stay paged in all of the time even though they really aren't very busy. >2) I noticed a significant drop in response time when I read news via rrn >compared to when I read news via nfs mounted /var/spool/news. Is this >general or was it just my case? One major reason for a decrease in response time while using nntp is that it seems to be a ack every packet type of system. When transmitting an article to the client the server sends a packet, waits for an ack from the client and then sends the next packet. There is no windowing that can be seen from monitoring the net. NFS uses windowing to increase trasmission speed. I have noticed this on every system I have used. >-- >Eiji Hirai @ Mathematics Dept., Swarthmore College, Swarthmore, PA 19081-1397 >hirai@cs.swarthmore.edu | hirai@swarthmr.bitnet | uunet!hirai%cs.swarthmore.edu >Copyright 1990 by Eiji Hirai. All Rights Reserved. Permission to reproduce or >quote explicitly denied except on Usenet. I don't speak for Swarthmore College. -- Too bad the universe doesn't run in a segmented environment with protected memory. -- Wiz from "Wizards Bane" by Rick Cook Brett McCoy | Kansas State University brtmac@maverick.ksu.ksu.edu | UseNet news manager.
stealth@caen.engin.umich.edu (Mike Pelletier) (07/24/90)
I would recommend using NNTP rather than NFS. Here, we've set up
a client directory with the portions of /usr/lib/news that would
be of interest to rn users, so the fact that you can only access
articles and the active file through nntp doesn't cause much trouble.
--
Michael V. Pelletier | "We live our lives with our hands on the
CAEN UseNet News Administrator | rear-view mirror, striving to get a better
Systems Group Programmer | view of the road behind us. Imagine what's
| possible if we look ahead and steer..."
aperez@cvbnet.UUCP (Arturo Perez x6739) (07/25/90)
About 2 years ago we used NFS to share news. I found that it took about 2-3 MINUTES to access each article read. Needless to say, at that rate it would take all day to read any significant amount of news. I once posted a query about it to the net (i.e. why is NFS taking so long) and everyone said "Your NFS is tuned wrong." Of course, since no one knew my network configuration and I don't know NFS I decided not to do anthing about it. I suspect it may have something to do with the 1/2 dozen or so gateways 'twixt me and the news server. Now we use NNTP. It takes me 1 hour (daily) to read the news from 60 newsgroups which average 10 messages each. Needless to say, we much prefer NNTP. We seem to average 10 nntpd running all day long, at a site with an /etc/hosts file that runs to ~2500. Some day I really must figure out who reads news... Please realize that this anecdotal evidence may be theoretically wrong. But NNTP works much better for us than NFS did. Arturo Perez ComputerVision, a division of Prime aperez@cvbnet.prime.com Too much information, like a bullet through my brain -- The Police