Will it run on multi-chip CMT?

April 15, 2008

Cooltst v3.0 is out, updated to assess workload suitability for single- and multi-chip CMT. When the UltraSPARC T1 was released, Cool Threads Selection Tool (cooltst) was developed to help gauge how well given workloads might run on the new chip which traded speed for throughput, allowing cooler, lower power, lower cost computing for many applications. But which applications? A single threaded application would tap just a tiny fraction of the 8 cores and 32 hardware threads of the UltraSPARC T1 processor.

Iguazu FallsMuch has changed since then. There is much empirical data showing various applications running well on CMT. The UltraSPARC T2 processor was released, increasing CMT power to 64 hardware threads. This processor also added dedicated floating point units per core so that, far from being relegated to a niche web server market, it claimed (and still holds) a high performance computing record.

Now UltraSPARC T2 Plus systems have been released, further extending CMT power to 2 chips, 8 cores per chip, 8 hardware threads per core – 128 virtual CPU’s in a 1RU box. Cooltst helps you assess how well your workload may tap that throughput potential. You can read about it and download it starting at sunsource.net.

There’s nothing magical about cooltst’s heuristics. You can make much the same assessment yourself using ordinary tools like ps (to look at the software threads) and cpustat (to look at instruction characteristics). All the source code is included so you can see what it’s doing. On Linux systems a loadable kernel module is included to measure instruction characteristics in place of Solaris’ built-in cpustat command. The output of cooltst is tabular data and a narrative description and  of your workload characteristics, and a bottom line recommendation.

Disclosure Statement:

SPEC and SPEComp are registered trademarks of Standard Performance Evaluation Corporation. Results are current as of 11/11/2007. Complete results may be found at the url referenced above or at http://www.spec.org/omp/results/ompm2001.html

My photo:

Iguazu Falls, on the border of Argentina and Brazil. It’s over twice as big as Niagara Falls in terms of water flow, because it covers such a wide area.


autonomous robots come to San Diego

April 11, 2008

The International Autonomous Robotics Competition is coming to San Diego in June, at the Del Mar Fairgrounds! Thanks to the San Diego Java Users Group and Wintriss Technical Schools, kids can compete in building and programming robots. The kids will use Sun SPOT which are – of course – the open source robot tool kits. They’ll program the robots in Java. Eric Arseneau writes all about the robot competition. I think my boy is a bit young to be writing software, but he’s taken me by surprise many times before. Contest or not, he wants me to get him a robot kit, which he thinks I must be able to pick up in the office any day.


SPEC does not certify results

April 9, 2008

Nothing is more fun than arguing with BM Seer. He usually helps me more than anyone in keeping everyone at Sun in compliance with SPEC’s fair use rules. But in a recent posting on SPECweb2005 for Sun SPARC Enterprise T5220 he refers to SPEC published results as "Certified." Actually as the official SPEC disclaimer spells out, "the contents of any SPEC reporting page are the
submitter’s responsibility. SPEC makes no warranties about the accuracy
or veracity of this data."

Most SPEC benchmark results can be used without SPEC review. They must comply with all the run and reporting rules, including the requirement for a full disclosure report. And their rules compliance can be challenged on the basis of the details in that report. There is a real value to readers, and hence to vendors, of publishing a result at spec.org. Such results are peer reviewed by other SPEC committee members including competitors, prior to publication. If a result is found to be not in compliance with the run rules it is not published, and the result cannot be used elsewhere either. However, passing this review is not a guarantee or certification that the result is accurate.

Instead of a paid independent audit process, SPEC relies on full disclosure and peer review to increase confidence in the reliability of results. From the details in the full disclosure report anyone should be able to reproduce the performance experiment and obtain substantially the same results. From time to time competitors will conduct such replication experiments on each others’ systems, and if they cannot get the same number they bring it to SPEC to either get some details of the test configuration that were erroneously left out of the full disclosure report, or to have the published result marked non-compliant. By this method SPEC dramatically lowers the cost of benchmarking, making it possible to have the thousands of results posted on spec.org, while keeping vendors honest by the fear of exposure and humiliation.

♪ Nuevo Flamenco and Reggaeton ♪

April 6, 2008

Jeff Tamarkin: "Simply labeling Barcelona‘s Ojos de Brujo a nuevo flamenco group is a
little like calling Disney World an amusement park: it’s way too
inadequate a description."
OdB is available at emusic.com. This music recommendation is specifically targeted to Alan Adamson, who enjoyed Bach por Flamenco. Coincidentally, Alan lives in Canada, and my favorite classical guitarist is Daniel Cox from Edmonton. Unfortunately as far as I can tell, Cox isn’t recording now. He produced several albums during the early days of mp3.com when its business model was to give all the music free, and hope a few people would, like me, also buy some CD’s of the same music in the hope that some of the money would make its way back to the artist and encourage them to keep playing.

Pick two is the Reggaeton Kings, also on emusic.com. Fortunate name for the band. If you’ve heard some Reggaeton and would like some more, where do you find it? In the old days you’d ask the knowledgeable clerk at the local record store. (Are you old enough to remember what a record store was?) The emusic categories are far too crude to find it. You could search for Don Omar, and along with some music find postings by purists sniffing that Don Omar isn’t real Reggaeton. Or you can just search for the word, and the Kings come up. I’m surely missing lots of Reggaeton from bands with less fortunate names, but that’s okay because the Kings are as the album cover says, Lo Mejor.


Forward to the 32-bit past

April 5, 2008

After wrestling with incompatibilities of 64-bit Linux for a while, I finally downgraded my home PC to 32-bit Ubuntu 7.10 (Gutsy). I found some nice and less nice workarounds, like running the Windows version of Firefox under Wine in order to get Flash to work. I hadn’t found a workaround for the Java browser plugin, or for Skype, and was considering a 32-bit chroot environment.

Finally one of those automatic updates decided it for me. You know the ones, the  messages offering later versions of software, critical security updates and recommended updates. Being able to just click OK to automatically be upgraded to the latest software is part of what makes Ubuntu so friendly. But this time it wasn’t so friendly. Something left my PC unable to boot to multi-user, unable to start networking, and unable to start graphics. I don’t know what because I didn’t keep the disk image around for a post-mortem. It was much much faster simply to blow away my root partition with a complete new OS installation. So while I was at it, I dropped down to 32-bit.

Lots of things started working, but some got worse. I had always had a problem on Gutsy that after suspend/resume the Ethernet driver would get a reversed MAC address, complain that it was invalid, and switch to a new eth instance with a random MAC. Of course this played havoc with my router trying to keep track of where my PC was in order to provide DNS. This problem occurs in the forcedeth driver, reverse engineered for the nForce chipset. Some people worked around the problem with limited success by adding commands in the suspend/resume scripts to stop and restart networking.

But now on 32-bit Gutsy it got worse. Upon resuming the screen stayed black, and since the network seemed to be down I couldn’t remotely login to find out what was wrong. I found lots of reports on the web about suspend/resume problems with the same error message in my .xsession-errors

Gtk-WARNING **: This process is currently running setuid or setgid.


This seems related to my NVIDIA GeForce 6150 LE graphics. Like many others who posted their experiences, the problem occurred for me both with the generic open source driver and with the Nvidia proprietary accelerated driver. One person mentioned a workaround by logging out, logging in to a failsafe X-terminal, and suspending manually from there.

Irony: The main reason I’m running Ubuntu instead of Solaris is that Solaris doesn’t yet have power management, and for a home PC, suspend and resume are essential. I’ve been eagerly watching the power mangement project Tesla at opensolaris.org, wondering why it’s taking so long. I guess like most things it’s easier to do, than it is to do right. By comparison, my PC’s when running Windows 98SE often fail to wake up at all, and those running Windows XP tend to wake up by themselves, unbidden. The only systems where suspend/resume always worked were Linspire and, of course, MacOS.

Neither workaround by itself would work for me, but putting them both together I end up with a clumsy workaround that lets me suspend/resume, and may possibly point the way towards a less cumbersome workaround.

  1. Disable networking via gnome panel
  2. Logout
  3. Select failsafe X-terminal session
  4. Login
  5. sudo /etc/acpi/sleep.sh
  6. (system sleeps)
  7. (normal wakeup by pressing ENTER)
  8. Logout
  9. Select normal gnome session
  10. Login
  11. Enable networking via gnome panel