[Gllug] UKUUG

[Chris Ball]

On Tue, Jul 03, 2001 at 09:06:36AM +0100, Jake Jellinek wrote:
> Anyway, the event was not the best organised thing in the world, and in
> desperate need of air conditioning, but the content was nearly all of an
> excellent quality and level.

Agreed in entirety. Even if I was helping out. *cough* :-)

> The IBM marketing guy (Nick Davis) was extremely embarrassing and there was
> a sad moment of realisation that much of IBM's involvement with Linux is
> done without much understanding of what is going on and apparently just in
> an attempt to make as much money as possible at any cost out of the latest
> buzzword. Of course we shouldn't mind that much I suppose, but I couldn't
> help thinking that many of us don't really care if Linux is the market
> leader in E-commerce management solutions and various other acronyms which
> even when explained just sounded like BS.

That's spot on. I was so annoyed by that talk. He kept talking
patronisingly about how we're all wonderful linux developers and we're
giving the world such great things and helping IBM do business too, and
then he started out on how we get special discounts on IBM's development
suites and access to their centres for helping us port our applications
to IBM's platforms, like Websphere. Missing the point somewhat, I felt.
Still, that shouldn't reflect on Richard Moore's (also from IBM) talks,
which you, like myself, probably saw one of (dprobes) at a GLLUG meet
a few months back. 

> Bo Thorsen from SuSE gave us a great insight into the process of porting
> Linux to a new architecture and had us all interested in the next generation
> AMD 64 bit x86 chip. He showed us his Simco hardware simulator (kind of like
> VMWARE with hardware specs you can design) and we even saw "ls" running in
> 64bit mode ;-)

I think it was 'simics' rather than Simco, but I don't have the CD
around to check the paper. I was fortunate to sit with Bo at dinner on
the first night - although it was a mixed blessing to have another
linux porter (for embedded systems) and a gcc developer on the same
table; I can recite several tens of reasons why gcc is bad, now. :) 

> [description of vsyscalls] 

I can't find anything wrong with your description of vsyscalls. The
slightly more descriptive version would be that an address is given
that can be executed instead of gettimeofday() - this executes a page
that is marked readable and executable at the user level, saving the
overhead on the context switch from user to kernel space. I think. :) 

I spoke with Andrea after the talk - vsyscalls are only implemented on
x86-64 at the moment, and I was curious as to why they weren't in the
main kernel stream. His reply was a little above me, but in a nutshell
they break binary compatibility, so it's best to wait for a new binary
spec. :-) It's a shame, though - the number of daemons (apache,
postgres) that could /really/ benefit from vgetpid() and vgettimeofday()
would have made it at least patch-worthy, I would've thought.   

> After this first day we had a banquet at Yang Sing, good quality chinese
> food in great company, a most enjoyable meal.

I'm going to be an AMD whore and point out that they gave us all
a Leatherman Micra at the meal, as well as sponsoring it. I hadn't seen
the Micras before, but the teeny screwdriver saved me from removing half
of my server last night, to get the PCI riser board out. Kudos to AMD
for knowing how to keep a bunch of linux developers happy. :-) 

> That sums up Friday (the first day). I'll leave Saturday and Sunday to
> someone else to describe, or come back to it later if anyone is actually
> interested in my waffling.

I'd love to hear some more of your views on the talks, and I'm sure the
UKUUG would very much appreciate the feedback, too. I'm writing up the
talks (in conjunction with another of the helpers) for the UKUUG
newsletter and anyone else interested in distributing them - I'll make
sure to post an URL once they've been okayed and distributed by UKUUG.

The quality of talks on the saturday was certainly as great as those
on the friday - while the friday was reserved for low-level kernel
talks, the saturday started a stream of desktop talks, including a
great presentation by Michael Meeks on the structure of Gnome, an
example of RAD with Gnome and Python from Christian Egli, and
Andrea's second talk on NUMA and O_DIRECT, which again had excellent
content.

Another enjoyable talk for me was the one Randy Appleton gave on Sunday
morning on his experiences of 'Benchmarking the Linux Kernel'. Randy
and the class he teaches in Michigan noted that the size of the kernel
has tripled since 2.0, and investigated whether that has sped up
operations like syscalls and i/o, or whether it had slowed them. He
benchmarked twenty or so kernels from 2.0 to 2.4, and showed graphs of
performance across the kernels. It was light-hearted but fascinating,
and he also described just how happy his classful of college students
were when their research ended up on the front page of Slashdot. :-)

Hope everyone enjoyed the weekend as much as I did,

~C.

-- 
Chris Ball.
chris at cpan.org || http://printf.net/

-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug