Mailinglist Archive: opensuse (6210 mails)

< Previous Next >
Re: [opensuse] Guru's RPM: YaST vs. Web
  • From: Eberhard Moenkeberg <emoenke@xxxxxxx>
  • Date: Sun, 9 Oct 2005 23:03:52 +0200 (CEST)
  • Message-id: <Pine.LNX.4.61.0510092120570.14586@xxxxxxxxxxxxxx>
Hi,

On Sun, 9 Oct 2005, Pascal Bleser wrote:
Randall R Schulz wrote:
Eberhard,
On Sunday 09 October 2005 11:22, Eberhard Moenkeberg wrote:

...
Is this just a glitch, or will the YaST repository remain behind
the other section of the Guru's repository?
... [Explanation of the woes of running a popular server] ...
Thanks for the information.

And I don't have much to add to what Eberhard already wrote ;)

Furthermore, to reduce the burden on the gwdg.de side, I generate the YaST2 and Red Carpet repository metadata on my server, so it is "just" uploaded and not computed on ftp4.gwdg.de

That means, in a purely theoretical world, that there shouldn't be any inconsistency between the YaST2 repository metadata and the availability of the packages (using YaST2's software manager or y2pmsh).

But if you meant an inconsistency between the newsfeed / list of latest packages on the website and the availability of packages in the YaST2 repositories, you are correct.
The rsync issues Eberhard wrote about are affecting my site quite dramatically, as the last 3 or 4 days my newest packages haven't been uploaded.
This afternoon an rsync job was able to complete successfully (by luck, it seems), so my queue is more or less empty now. But we don't know whether it'll keep on working or not (probably not).

The list of latest packages is just a file where I add the entries "manually". Hence, if I announce a package there but the rsync doesn't fetch the package properly, it is inconsistent.

Sorry for the inconvenience, but as Eberhard explained, it's hell for the ftp*.gwdg.de servers at the moment.

So, change ftp to ftp4 in your YaST installation source configuration
if you run into inconsistencies, but be aware that you extend my pain
each time you use it. ;-))

Death by a thousand cuts, eh?
Sorry. I will exercise patience.
Is there a mirror for the YaST-compatible Guru's RPM Repository? Surely
some of the popular mirrors would make enough space available, wouldn't
they?

Well, unfortunately not. If someone would like to stand up and mirror my packages, I'd be the first one to be happy about it, but I didn't have any offers for that (yet?).
If you'd like to support me in what I'm doing, and especially the many users who download my packages, *and* you have bandwidth and at least 3 GB of space to host it, let me know :)

Normally a few servers are also mirroring the suser-* directories of ftp.gwdg.de, so there should be a few out there, but I don't have a list.

If you run into a mirror that does, please send me an e-mail (guru AT unixtech DOT be) to let me know, I'll publish that information on my website.

Again, I'm really sorry for the inconvenience, but I'm afraid there's not much to do about it at the moment. Seems like 10.0 draws a lot of attention and... downloads ;)

Yes, this is the best "net" effort for SUSE ever. No doubt. So, my one "whining eye" is accompanied by the other - laughing as never before. This lousy bottleneck is making success, success like never before and never thought of. ;-))

And seeing the bottlenecks (at the servers, and at the distribution scheme, too), some good ideas for a better future are coming up.

W. I. Lenin once said: "the practice is the proof of the truth" (in german: "die Praxis ist das Kriterium der Wahrheit"), and this will be my motor:

First, I am struggling for a new server with 32 GB RAM, to have enough buffer cache even if the real disk I/O is suffering (as it will always be in these /. situations).
No real chance from the GWDG budget, but maybe one of my external ftp.gwdg.de protegees will find a sponsor.
I would love to use a quad Opteron like Newisys 4300-E (the original "reference design" for quad AMD Opterons and still on top, even with the new DualCore Opterons).
But I guess Sanmina/Newisys does not care about ftp.gwdg.de and maybe even does not even have any picture about, so I need some ambassador for this idea more directly towards AMD.
AMD has the A for "advanced", not for "America", and Dresden ("saxony valley" in Germany) is the Opteron production center, and the german government has declared to help AMD to convince Intel's bad practices against AMD in Germany (read: Europe). You see. I am dreaming especially about this hope...
HP (the Opteron server market leader) already has sponsored kernel.org with two Proliant 585, so they have done their duty, I guess they guess.

But second: I will try to get a better "real" disk I/O soon, by using two IDE-Raid arrays instaed of a single one, concatenated with LVM in striping mode. This can theoretically double the real disk I/O rate, but in practice noone would pay for a second IDE raid array currently, so I have to test with two "older generation" ones; but hopefully I will see a positive result which is good enough to dream upon.

And third we can try the trick which had already success on ftp.gwdg.de with the i386 ISOs: to help that the most important files for SUSE-10.0-OSS will remain in buffer cache once they got there, by provocing cache hits again and again, tight enough in time.
This is the idea behind "minute 0 of every hour for OpenSUSE" - please participate to make it work. ;-))

Cheers -e
--
Eberhard Moenkeberg (emoenke@xxxxxxx, em@xxxxxxx)

< Previous Next >