Standard Wiki Info

• Name: Pat Murphy
• Email: pmurphy+almatwiki@nrao.edu
• Yahoo, (G)AIM: LinuxGuyInVa (here's why I don't use skype any more)
• Phone: +1-434-296-0372
• Company Name: NRAO
• Company URL: http://www.nrao.edu/
• Location: NRAOCharlottesvilleOffice
• Country: USA
• ALMA involvement: I participate in the COMMS group.
• Comment: I worked in the ALMA computing IPT (Software Engineering) from January 2006 through August 2007. See also my NRAO internal and external wiki pages. That's me in the corner.

Software Engineering

• Makefile Analysis: Some interesting and possibly arcane results:
  • Running builds with ACS accessed via NFS causes a significant performance hit.
    • Try to use a local ACS instance if possible.
  • Overhead of the make infrastructure is no more than 20-30%.
  • Alternatives to make may be compelling, but the cost of a change is too high at this late stage in the project.
• Subversion Timing: Performed timing tests on a Subversion (svn) server.
  • Conclusions: svn is not really any slower than cvs (but it's not faster either).
  • Others report much worse performance if the target directory is not local, but NFS mounted. This bears further investigation, but could be a real pandora's box (think: NFS options, speeds, network configuration, automount issues...). A comparison of the two servers on a NFS file system would be instructive...
• SPR ALMASW2006144: detect unresolved symbols in shared libraries at compile time, not run time.
  • Tools that can help: readelf (to find unresolved references in a library).
  • Resolution: a new script acsMakeCheckUnresolvedSymbols which uses ld on a shared library (.so) to find unresolved symbols. This script gets called via the standard acsMake file at the appropriate time in the dependency checking.
  • Reference: LinkingOnLinux
• BonnieBenchmarks (NFS):
  • I used the Bonnie benchmark suite to compare NFS performance on various systems, specifically as implemented in the ATF STE.
  • Bonnie gives results for three types of I/O, with up to three modes (see below).
  • Conclusions:
    • NFS write performance from a desktop at NRAO-CV-NTC to a filer at NRAO-CV-ER (3km) was 100 times slower than local disk.
    • Full results in this OpenDocument spreadsheet. These include numbers for sequential input and output on both character and block I/O, rewrites for sequential output, and random seek I/O. Sequential per-character I/O probably simulates the action of writing to log files best, whereas block sequential I/O is a good measure of bulk data writing speed. The sequential input block I/O figures were hampered by the very aggressive Linux disk cacheing. Random I/O is likely a good measure of data reduction performance (including pipelines).

Computers & Communications ("COMMS")

• I run the monthly COMMs meetings, coordinating purchases of hardware in the near term (routers, switches, computers; basically, items the Computing IPT has agreed to supply as specified in various Interface Control Documents).
• I helped to fill the gap between EDM documents (long time constant for formal change) and reality with a couple of wiki topics. Case in point is the ALMA Computing Standards document:
  • HardwareStandards
  • SoftwareStandards
• I helped the process of deriving other standards and conventions (these are ongoing):
  • DNS Naming Convention
  • ALMA IP Addressing
• Hardware at OSF : StefanoTurolla did most of the work here but I tried to help too. I started a (visio) diagram showing the computer and network gear that ALMA Computing has committed to buy, (attached to HardwareOSF). ChristianSaldias now maintains that diagram, and another one specifying hardware for the AOS (still under development).
• HighAltitudeGear shows tentative specifications for both laptops and servers that need to operate at elevations such as those found at the AOS building.
• I spent much time in early and mid 2006 editing a very large word format document called the Computing Communications and Information Services Design document (ALMA-70.05.00.00-005-A-SPE); this was submitted to review by GianniRaffi and BrianGlendenning, and approved. A second revision occured during 2007, with major input from FlorianHeissenhuber. ChristianSaldias will be handling any future updates to this document.

Charlottesville STE

• I helped co-ordinate the acquisition and setup of a new STE for NRAO/Charlottesville.
• The STE is located in the communications room of the NTC-A building on the first floor.
• The CV STE was used in February 2007 for a major Refactoring effort.
• This STE does not (yet) have a 3-monitor operator console.
• A second gcn (diskless) server was added in August 2007.
• See MurphysVNCRecipe for how I recommend using VNC with a ssh tunnel.

Older Items

• Ethernet interfaces that aren't there: these cause delays in bootup on some clients. I tried a facility called ifplugd, making it operational in a RHEL environment. This (ifplugd) is small and lightweight, and automatically brings up specified interfaces when a carrier is detected (i.e., when you plug the cable in). See the IFPlugD topic for more details on this little project.
  (Now that I run Ubuntu Linux most of the time, and it incorporates the same technology as ifplugd, this is somewhat moot to me personally).
  Conclusion: nice for personal use, probably as much a liability as a benefit otherwise.
• LinkingOnLinux is a summary of issues affecting the linking (loading) process and shared libraries.

Calendar

Those of you with access to NRAO's internal wiki can look here for my NRAO calendar. I also have a google calendar; contact me if you'd like me to share that with you.

Wiki Settings

Personal Preferences (details in TWikiVariables)

• Horizontal size of text edit box (was 70):
  • Set EDITBOXWIDTH = 76
• Vertical size of text edit box
  • Set EDITBOXHEIGHT = 34
• Style of text edit box. width: 99% for full window width (default), width: auto to disable.
  • Set EDITBOXSTYLE = width: 99%
• Only yours truly and wiki deities can edit this page
  • Set ALLOWTOPICCHANGE = PatMurphy

Attachments

If there's nothing here, look on the HardwareStandards topic (at the end), or at the BonnieBenchmarks topic.

• Saber, Hale-Bopp, and me.: