Messages in this thread | | | Subject | List sluggishness, and occasional deletion of people from the list | From | Matti E Aarnio <> | Date | Tue, 4 Jun 1996 12:56:09 +0300 (EET DST) |
| |
Propagation delays do vary per direction..
> From owner-linux-kernel-outgoing@vger.rutgers.edu Tue Jun 4 12:02:57 1996 > To: submit-linux-dev-kernel@ratatosk.yggdrasil.com > Path: cortex.corpus.uni-muenster.de!not-for-mail > From: gutschk@uni-muenster.de (Markus Gutschke) > Newsgroups: linux.dev.kernel > Subject: This list is far too slow (Was: Re: 1.3.97 oops) > Date: 3 Jun 1996 19:58:39 +0200
So this came thru in less than 24 hours -- from Germany via Yggdrasil thru VGER, and finally to me.
Earlier sluggishness samples:
> > MAA05711; Mon, 20 May 1996 12:46:58 -0400 > > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ > > Received: by vger.rutgers.edu id <105941-12788>; Wed, 8 May 1996 14:13:37 -0500 > > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ > > That's 12 days from one machine to the next in the same domain!
> Your message arrived here on June 3rd, whereas the header says that > you sent it on May 20th; so that is another two weeks. The 12 day > delay seems to be quite typical :-( BTW, I am using the bi-directional > gate-way to the newsgroup linux.dev.kernel and submit articles via > submit-linux-dev-kernel@ratatosk.yggdrasil.com. I switched to this > after repeatedly being unsubscribed from the list. I would really > appreciate if somebody figured out why the turn-around time is *this* > long.
Apparently the feed to *.com is VERY SLOW -- I suspect it is being fed thru some sendmail, which is dying under the load (while vger hums along with its zmailer ..)
There are multiple problems -- vger.rutgers.edu is VERY LOW POWERED MACHINE, and thus it can't deliver to the whole world all by itself. It needs to push some 200 000 emails every day, and that is more than even ZMailer can do at the given hardware via direct SMTP sessions. (I have benchmarked ZM to break the "one million messages per day"- mark, but it did require very powerfull machine -- like a magnitude more powerfull than VGER.)
To be able to handle email to the world the VGER thus needs re-exploders, which usually are configured per the top-level domain -- *.de is routed to nic.funet.fi, *.com is routed to work.nether.net, etc. (Most of the Europe goes to nic.funet.fi, except NO and FR. Most of the southern part of Africa goes to paradigm.co.za, and most of the central/southern America go to Mexico)
So, direct email from VGER's lists to Germany (*.de) goes to nic.funet.fi, which does actual delivery. Email to the relay at the Yggdrasil.com goes via *.com exploder, and it apparently takes some time...
I added now an explicite route for yggdrasil, which perhaps speeds up the delivery to linux.* -lists via their gateway.
However the use of top-level domain is NOT a sure way to do optimal routeing, quite contrary. There are many Alternet customers out in Germany (with addresses of *.de) who are unreachable from Finland, and thus I can't deliver them email via nic.funet.fi!
Now adding exceptions at VGER for them is suboptimal thing to do, as they can't reach ftp.funet.fi anyway, thus I have asked Alternet technical contact for a fix for them, but so far have gotten no reply...
We do need email exploders to help vger -- or the lists will collapse. I am again calling for German, and UK sites that are willing to act as exploders -- you may need ZMailer in case your sendmail (or smail) dies under the load, though. I can even use multiple volunatry systems to share the load. ( Normally I would let VGERs real admins do the organizing job, but David Miller went to SGI for some project ... ) ( Ask me about the ZMailer )
> Markus > -- > Markus Gutschke Internet: gutschk@math.uni-muenster.de
/Matti Aarnio <mea@nic.funet.fi> <mea@utu.fi>
| |