You are viewing illiterat

James Antill - Understanding why YUM seems "slow", and some instant solutions

May. 21st, 2008

07:02 am - Understanding why YUM seems "slow", and some instant solutions

Previous Entry Add to Memories Share Next Entry

Comments:

From:(Anonymous)
Date:July 4th, 2008 09:45 am (UTC)

filelists.sqlite.bz2

(Link)
The filelists metadata are loaded too often. This is the real pain. They are big. Bigger than many updates. They are required for only few dependencies and just to make exotic packagers happy.
[User Picture]
From:illiterat
Date:July 7th, 2008 02:58 pm (UTC)

Re: filelists.sqlite.bz2

(Link)

If you wish to argue that file dependencies should be removed from Fedora, then you'll have to bring that up with Fedora ... all yum can do is load the data it requires, when it requires it. Although, note that you can change metadata_expire to control how often yum will try and refresh any of it's metadata.

From:(Anonymous)
Date:October 16th, 2008 10:32 pm (UTC)

Re: filelists.sqlite.bz2

(Link)
Downloading the meta data files every time there is an update in the repository (possibly as little as one rpm among thousands) is inefficient. Over my dial-up connection (there are still many of us out here) it takes almost two hours to download all the meta data.

What is needed is a new organization of the meta data on the server so that smaller downloads suffice for installation of a package and its dependencies. Even a full update shouldn't require downloading all the meta data because most systems have only a small subset of all available packages installed.

Ideally, this new data organization would be a joint effort between the createrepo, apt-rpm, smartpm, red carpet and yum projects (the projects using the current format according to the createrepo project page).

None the less, I prefer yum to rpm and I am looking forward to the improvements in the new release, whenever RHEL/CentOS catch up with it.

[User Picture]
From:illiterat
Date:October 17th, 2008 04:45 am (UTC)

Re: filelists.sqlite.bz2

(Link)

Well yum is almost certainly the best for constrained metadata download bandwidth, due to the fact it will (by default) only download the file or changelog data when yum needs it to do some operation (most of the other options want to download everything at once). So in the simple case of installing a single package, it's very likely that you can just download the "primary" metadata file.

But, yeh, even so that can still be relatively large if you want to install a small package over a modem. yum-presto will be available/integrated in Fedora 10, which I think is a step in the direction that'll help (it provides/uses delta information).

I don't think we want to standardise anything right now, because I don't think anyone knows what the final solution is going to look like (repo. metadata was only standardised after a few different people had tried different things, and the problem was fairly well understood -- and it's still far from "perfect"). But saying that, we do keep in touch with the SuSE/zypp developers.