John DeHart Washington University, Applied Research Lab jddarl'wustl'edu http:www'arl'wustl'eduarl

1
John DeHartWashington University, Applied
Research Labjdd_at_arl.wustl.eduhttp//www.arl.wust
l.edu/arl/
MSR Tutorial Notes from Day 1, MSR Code
Release and Future Efforts
2
Program Notes

• Our Grant ends this June
• One last Workshop in early June 2002.
• Any suggestions for what it should be like?
• Summary talks from Kits Participants
• Summary talks from us
• Tutorials
• What would be most useful for you?

3
Tutorial Notes

• Slides will be available on the Kits Web pg.
• http//www.arl.wustl.edu/gigabitkits/kits.html
• look in Workshop 1/2002 link
• the slides arent there yet, try at the end of
  the week.
• TUTORIAL directory
• will be included in the CVS release of msr
• in its original form
• if you have anything you did that you want to
  save, ftp it back to your site today.

4
Notes from Day 1 Exercises

• Download problems
• Download fails regularly right after reset
• POSSIBLE REASON FOUND
• There was a problem with reset in Stage1 of the
  AAL5 downloader. It is being fixed now and will
  be available in the next release (00B)
• check /proc/net/atm/wuapic for driver state
• possibly a timing problem
• possibly one SPC taking longer to reset (seems
  unlikely)
• Download returns lots of errors but completes and
  seems to be ok
• POSSIBLE REASON FOUND
• Turns out that at least one of the switches
  involved in the Tutorial was generating parity
  errors. So, there would have been dropped cells.
  The good news is that AAL5_download was able to
  robustly function in this environment!
• check /proc/net/atm/wuapic for driver state
• possibly a pacing problem and some cells are lost
  on the download to SPC
• fix by slowing down the VCI pacing for downloads

5
Notes from Day 1 Exercises (contd)

• AAL5Generator
• deak would not run it. malloc error
• FIXED notest gave path to Linux executable
  instead of NetBSD executable.
• when you leave a generator running and then
  re-run the download, the kernel crashes during
  boot
• During boot when we get to the point of turning
  on the VCIs the kernel gets swamped with
  interrupts and there is probably something
  uninitialized causing us to crash.
• We are looking at leaving the data VCIs turned
  off until the initialization phase is run.

6
Notes from Day 1 Exercises (contd)

• newGBNSC
• Burns up the CPU when you give it too small a
  sampling period.
• at least I think this was the cause of the
  problem
• we should put a bound on this in the newGBNSC
• javaGUI
• drag and drop
• switch between cells per second and Mb/sec
• Zoom in with selection
• have to be inside the axes to draw the selection
  box
• then click in the box to make the selection
• Kernel hangs around time of starting init
• no clues on this one yet.
• anything else??

7
Code Release

• Release via CVS
• let me know if your group does not have a cvs
  account set up with us
• should I resend the CVS usage instructions?
• Same wu_arl source tree
• additional sub-directories to be included
• New release to be announced shortly.
• well email out the release tag
• Schedule
• We have a few bugs to fix based on some things we
  have learned in the last couple of days.
• Release should occur by the end of this week
• Watch for email to the kits mailing list!!!

8
Near Term Future Efforts

• SPC II
• board might go out to fab next week.
• FPX Integration
• MSR CP Expansion
• Elimination of dependency on shell scripts
• Routing protocol support
• Distributed Queueing
• another round of testing
• DRR and WV
• demonstration of the WaveVideo active application
  on the MSR
• Support dynamic and remote loading of Plugins

9
Near Term Future Efforts

• Gigabit Ethernet Interface
• board probably out to fab in early February.
• Light weight flow setup protocol
• Binding of plugins to exact match filters
• needs to be tested and verified
• Simple timeout mechanism for flushing cached flow
  and route entries.
• Restructure buffer management so we can source
  and sink packets within plugins.
• Wow, were going to be busy