[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: anaconda- performance ...

John Summerfield wrote:

Alan Milligan wrote:
Just an FYI, but I've just built and deployed essentially an FC6 +
Updates + loads of Zope and Plone using the latest devel anaconda and

The performance was glacial across both an AMD x86 desktop (admittedly
with only 256MB RAM - the rest was purloined by my x86_64 build server),
and my Sony laptop with 740MB RAM.

The installer is a DVD with circa 1.1 GB image and 900 packages, the
upgrade involved circa 750 of these.

The act of resolving dependencies for install took around 15 minutes, an
hour to install the 750 packages, and the best part of 15 minutes to
cleanup - total time circa 1.5 hours for a simple upgrade.


I do hate to project the outcome of a 4GB DVD installer ...

Thoughts anyone??

Just to test something....

Create an archive on disk and do an http install to match the one you've done. One hopes you're using kickstart, and if not, that you will:-)
Strangely, I was experiencing real problems setting up the NIC to do http installs. But these don't need an archive disk/iso image they're expecting a file layout as per the iso. However, I'm serving these up with Zope which isn't necessarily the quickest thing on the planet, and sticking Squid in front of it isn't going to do much for one-off requests...
The time taken for that is useful, but not the point. Analyse the webserver logs and use the information there to create a file for mkisofs so you can specify the file order.

I don't quite understand why I should do this. I already have a perfectly good pkgorder.txt and know what the file order will be ...

Create a new dvd image, and install from that.

Report on the difference, I've long been curious about this, but never enough to do some kind of benchmark.

I'd be interested to know the time for each of the three; wristwatch measurement is fine.

I actually think that yum's causing quite a few of these problems whilst we await a higher performance version to land. I'm still surprised at the length of time taken to unpack the RPM packages - I can't imagine why there'd be major changes in the RPM libraries/Python bindings that could cause this.

I'm a couple of days away from doing any more testing unfortunately. While I can't retest these upgrades, I will post some qemu install results - hopefully I can get the kemu accelerator working as it's even more glacial than the above ;)


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]