AMD 64-bit Opterons brought to O'Reilly FOO camp

by Andy Oram



Two AMD employees came by the Friends Of O'Reilly (FOO) camp yesterday
to build two systems based on their 64-bit Opteron processor, running
SuSE Linux. Our lead database programmer, Tim Allwine, is trying out
one of these in the hope it will vastly speed up the generation of
statistics regarding book sales. With the 64-bit addressing and eight
gigabytes RAM, he can instruct MySQL to slurp all the data into memory
and go from there. I also caught some information on GNOME, security,
and the Segway--all in one day.



Opteron on





Rich Brunner, AMD Fellow, introduced their chip in a brief
presentation. Some of the interesting parameters are:






  • The default address size is 64 bits, but the default data size is 32
    bits, which helps keep code small.



  • Doubling the number as well as the size of registers offers a major
    speed-up for 64-bit applications.



  • An integrated memory controller halves the latency of memory accesses.







Memory hangs directly off of each processor, whereas in earlier chips
it had to go through the same bridge as I/O. Furthermore, each
processor can support three of AMD's HyperTransport interconnects,
which makes it easy to connect multiple processors. Accessing another
processor's memory takes only a bit longer than accessing attached
memory, making NUMA less of a burden.



Other events





All day yesterday I talked to people I've been hoping to meet and
others I was glad to discover. Two authors who had been told not to
come unless they finished their book showed up at a quarter to
midnight. I rode a Segway, finding that I could do some fairly
sophisticated maneuvering right away.




Nat Friedman of Ximian presented his nifty search tool Dashboard,
which he had shown at the O'Reilly Open Source conference last July,
but which now sports a couple new features like an index for
everything on the desktop. He is leaving tomorrow for India, where he
will meet with a large number of programmers employed by Novell, the
company that bought Ximian recently. He will recruit 30 to 60 of these
programmers to work on GNOME and help them learn the social
conventions of working in a free software environment. Meanwhile,
Ximian will make GNOME more robust and get several important
subsystems in shape, such as plug-n-play and printing through CUPS.
One of their upcoming projects is a blogging tool. They hope soon to
make Mono compliant with the .NET 1.0 spec, or close to compliant.




The hall was full for Bernie Krause, who is billed as a "bioacoustics
author." He has spent several decades making sound recordings of
natural habitats, and has thus inevitably become the chronicler of
natural habitats' extinction. It really brings home the impact of
heedless policies to hear recordings of an environment's richness
before selective logging, and the poverty of sound a few years after
that logging. Even noise from jet planes can disrupt the ability of
species to carry on the business of living.




Author and telecom expert Brian McConnell described a phone system
he's marketing that allows cellular phone users to dictate email
messages. The operator-assisted service delivers them to people's
email accounts as attached MP3 files. He expounded on the value of
mixing humans and machines, using the strengths of each in the system.
He also mentioned that phone companies should have saved the billions
they invested in 3G and developed slow but useful technologies such as
GPRS instead.




I learned from a security researcher why not so many of our security
problems would go away if all common servers were coded in Java
instead of C. It seems that we're always inventing new, cool ways to
intertwine systems in more and more complex ways, each of which opens
up new attacks. Typically, security experts can find five exploitable
errors in each thousand lines of C code, while for Java code the error
rate drops to one in each thousand lines. A thousand lines of code is
not a lot nowadays.



Wrap-up





Leaving aside the beautiful setting and the fascinating attendees, a
couple people here put their fingers on what they love about FOO camp:
they say it has the informal get-togethers and impromptu demos they
always loved about conferences without the formal presentations and
keynotes they always hated. FOO camp takes the frosting from
conferences and leaves out the angelfood cake.




Many people took off last night, but those who stayed have continued
at almost the same intensive pace; many sessions are attracting rooms
full of active participants. I have to admit that my session on doing
books under open licenses (offered by popular demand, strangely
enough) attracted only one person, but we had a fascinating
hour-and-a-half long discussion spanning scads of topics.




Hardly anything has been posted on the Internet about the camp. I
think most people don't know what to make of it yet. But in a few
months one may start hearing people say, "This new project came out of
something I was chatting to folks about at FOO camp."



I also covered FOO camp in
yesterday's weblog.