Page History
- 2012-Nov-13 16:47 adesmet
- 2011-Aug-24 14:14 adesmet
- 2010-Mar-30 13:48 adesmet
- 2010-Mar-30 13:30 adesmet
- 2009-Nov-23 11:24 vmathew
- 2009-Oct-21 12:20 vmathew
- 2009-Oct-14 21:16 vmathew
- 2009-Oct-09 12:45 vmathew
- 2009-Oct-08 14:57 vmathew
- 2009-Sep-29 10:58 adesmet
- 2009-Sep-29 10:57 adesmet
- 2009-Sep-21 15:26 tannenba
- 2009-Sep-17 11:06 tannenba
- 2009-Sep-15 13:34 jfrey
- 2009-Aug-17 10:33 gthain
- 2009-Aug-04 12:03 adesmet
- 2009-Jul-13 09:27 gthain
- 2009-Jul-07 16:45 tannenba
- 2009-Jun-23 13:13 adesmet
- 2009-May-28 13:34 tannenba
- 2009-May-28 13:30 burnett
- 2009-May-28 13:26 burnett
- 2009-May-20 13:08 tannenba
- 2009-May-19 15:20 tannenba
- 2009-Feb-23 14:17 psilord
- 2009-Feb-19 10:44 psilord
- 2009-Jan-26 11:39 tannenba
This page contains an unsorted list of brainstorm ideas that could be good candidates for student hourly work.
- A program written to show how the glibc keeps open the socket to the nscd so Ulrich can fix it. Basically, it needs some twiddling, using strace, waiting around along time, a select() loop waiting for ages, showing the connection doesn't go away, that sort of thing.
- Go through all the Coverity bugs and fix them (roughly 1200 bugs at last count)
- Add support for standard universe to the new starter/shadow, then dump the old one. This will remove all kinds of inconsistencies between vanilla and standard universe, and speed up future development, as we won't have to add code in multiple places.
- There's a number of cases where condor_q -analyze doesn't know why jobs can't run. (negotiation cycle hasn't happened yet, job retirement in progress, parallel universe jobs, etc.) It would be nice to add smarts to -analyze to catch these cases.
- Add stress tests to the Condor test suite. It would be great if we had tests that stressed the individual daemons, without trying to run jobs through the system. This way, we'd find memory leaks and other embarrassing problems before out users did.
- Currently, CondorView is difficult to install and maintain. This is because it is written in perl, java, javascript and C++. I believe we could implement something much better in javascript + SOAP to the collector, and have it installed by default with Condor. This would just work without any tinkering around, installing apache, downloading jar files or anything other than just a normal Condor install.
- Make metronome scalable. There's been a bunch of times where a test fails intermittently. We've got plenty of machines. I'd love to be able to submit 100 test runs overnight and look at the results in the morning. Currently, metronome can't do this.
- Add more effort to writing tests, incl unit tests. I.e. more than just Bill.
- Job Sandbox transfer plugins: HDFS, BitTorrent.
- Work on the Stork supplement work.
- Wrap new classads w/ old classad API, and/or make a shared library.
- Get GCB working on Windows (at least the client library), and/or explore use of OpenVPN or SOCKS in place of GCB.
- Explore various Linux checkpoint solutions out there (kernel, etc), add job wrapper support. Ditto for Parrot w/ Condor.
- Secure setup by default, with use of MiniCA and/or online CA.
- Super CHTC turnkey package for installation on Windoze machines around campus, using combo of CoLinux + OpenVPN + Condor.
- Quill to MySQL (or SQLite?)
- Test Condor w/ out of disk space.
- defaults for param(), ala firefox...
- Use the streaming code already present and allow screen to connect to a running job. Or, start jobs under a VNC server process.
- Get a distcc setup going, put some abstractions into Metronome so you can automatically use it
- There are some memory-testing scripts running on GLOW using Hawkeye/startd cron. Include those in the default install of Condor
- Do a clipped port to Solaris 10/x86
- Rebase Todd's 739 project to 7.3.0
- Prototype a 'condor_job_backup' command (not sure you can finish this in a few weeks)
- a .NET/Managed code universe. Run the same job on the CLR on Win32 or on Mono (or, think real hard about how to abstract the Java universe, .Net universe, Python universe, etc)
- Replace the checkpoint server protocol with HTTP GETs and PUTs
- Create a couple of "Condor Technical Articles", in the 5-10 page range. Some ideas:
- A document exploring all of the issues involved in running Condor in a root squash environment. Document all of the different user ids Condor uses and what it does while it is in each of those UIDs
- A document that explores Condor and DNS - figure out all of the places Condor requires a working DNS, look at where it doesn't need it, and what you can override. Don't be afraid to start at square one with these documents, and explain what root squash is or what reverse DNS is.