1. 25 Mar, 2003 1 commit
  2. 24 Mar, 2003 10 commits
  3. 23 Mar, 2003 6 commits
  4. 22 Mar, 2003 4 commits
    • Mac Newbold's avatar
      Grab a batch at a time instead of a single node per loop iteration. · 4a34327a
      Mac Newbold authored
      Scaling and speed now depends primarily on os_load (and indirectly,
      node_reboot). The time a batch spends in the reload_daemon code appears to
      be <1s per node now, instead of taking 30s per node to grab, setup, and
      reboot.
      
      Also, finally remove the "obsolete section" that's been sitting in there
      for a long time. This was the part that did netdisk reloads, and has
      already been neutered out of the code path for several months at least.
      4a34327a
    • Leigh B. Stoller's avatar
      Add three new control mechanisms to the front end to control the use · 62b4edc9
      Leigh B. Stoller authored
      of emulated and linkdelays on a per-link/lan basis (although you
      cannot make a lan emulated at this time).
      
      	tb-set-link-uselinkdelay $link0 1
      	tb-set-lan-uselinkdelay $lan0 1
      	tb-set-link-emulated $link0 1
      
      Added slots to the virt_lans table, used in assign_wrapper to control
      these features on a per link/lan basis. Note, the global options are
      still available to prefer/force linkdelays for an entire experiment.
      When setting uselinkdelays on a per link/basis, if the link would not
      otherwise be shaped, you are requesting that a linkdelay be inserted
      anyway (in otherwords, its the equiv of tb-set-forcelinkdelays on an
      individual basis).
      62b4edc9
    • Mac Newbold's avatar
      Fix small bug with clearing swap request counter when an expt appears to · 4cc199f1
      Mac Newbold authored
      have gone active again. If you gave it a higher time than the default, and
      an idle expt hadn't been idle that long, it would clear the counter when
      it shouldn't, making idle expts look like they may have gone active again.
      (It didn't ever change any slothd data or cause any activity, you'd just
      "forget" that you'd already been sending swap requests. And it only had an
      effect when you picked times _higher_ than the default idle time.)
      4cc199f1
    • Mac Newbold's avatar
      Take the default idleness threshold from 6 hours down to 2 hours. Main · 2a977df3
      Mac Newbold authored
      reason is to keep things moving during the high SOSP demand, but it may
      stay there or somewhere similar when we deploy the new slothd next week.
      2a977df3
  5. 21 Mar, 2003 9 commits
    • Kirk Webb's avatar
      Changes to startup scripts to fire up the new slothd (uses tmcc to grab · 25bf7588
      Kirk Webb authored
      parameters). Install target update as well.
      25bf7588
    • Kirk Webb's avatar
      Well, here it is: The new slothd and sdcollectd stuff. · a337f7b9
      Kirk Webb authored
      Several things have changed; here is an outline:
      
       - "Thresholds" added to slothd
      	These are checked at each iteration against current idle data
              to determine whether or not to set bits in the "active" vector.
      	- LAST_TTY_ACT (now) > LAST_TTY_ACT (previous)
              - LOAD > <provided value>
              - EXPER_PKT_COUNT(now) - EXPER_PKT_COUNT(prev) > <provided value>
      	- CNTRL_PKT_COUNT(now) - CNTRL_PKT_COUNT(prev) > <provided value>
      	- Values passed in via command line: do ./slothd -h for help
       - "Aggressive" mode added to slothd
      	When a regular interval's worth of time has passed, and no activity
              has been noted, slothd goes into "aggressive" mode.  In this mode,
      	slothd changes its sleep interval, checking for activity much more
      	frequently (still reporting every <reg_interval>) until activity
      	is noted.  This is based on the threshold logic above.  When activity
      	is seen, slothd sends a report, and goes back to "regular" reporting.
      	- params passed via command line for regular, and aggressive intervals.
       - sdcollectd updates data in new "node_activity" table
      	- Updates timestamps based on activity bits in received slothd packets
      	- No new rows are added.
       - Option added to sdcollectd to insert into "old" node_idlestats and
      	iface_counters tables
         	- This is not the default (-o)
       - New sdcollectd still accepts old client reports
      	- but does not update the node_activity table with them.
       - Misc.
      	- Code cleanup
      	- Scan "active terminals only" option & code removed from slothd
      	- Always send report immediately on startup (no more option)
      	- Randomize time of second report to mitigate synchronization
      	- Send report on shutdown (receipt of INT, TERM, or QUIT)
      a337f7b9
    • Leigh B. Stoller's avatar
      Fix up the emulated link vlan stuff; build a supervlan of nodeports · 025728c5
      Leigh B. Stoller authored
      when doing emulated links so that ports do not end up in more than
      one vlan.
      025728c5
    • Russell Daniel Christensen's avatar
      We now skip the swap file and hibernate file on NTFS file systems! · fb8aa621
      Russell Daniel Christensen authored
      Minor changes: Updated code to work with the new libntfs library, and made the NTFS section of the code adhere to Leigh's 80 column decree.
      fb8aa621
    • Robert Ricci's avatar
      Make fixing plink endpoints no longer the default - needed for · d70e549a
      Robert Ricci authored
      virtual links.
      
      There will need to be a way to specify this on a per-vlink basis
      for jails.
      d70e549a
    • Robert Ricci's avatar
      Added a few more changes: · aad09909
      Robert Ricci authored
      * Make php.ini file
      
      * Edit ssh_config to prefer protocol version 1
      
      * Make /usr/testbed/tftpboot, and link /tftpboot to it
      
      * Make mountpoints for places we mount ops
      
      * Make sure that the DNS names for ops resolve before we put them
        into /etc/fstab
      
      * Chase emulab-boss port version
      aad09909
    • Robert Ricci's avatar
      Add a few more dependences, bump version. · fd774928
      Robert Ricci authored
      fd774928
    • Robert Ricci's avatar
      3a772ef5
    • Leigh B. Stoller's avatar
      Fix up widearea.html to equalize importance of RON and PCWA. · f0936bf6
      Leigh B. Stoller authored
      Change approveproject.php3 stuff so that approving ron includes pcwa.
      f0936bf6
  6. 20 Mar, 2003 10 commits
    • Kirk Webb's avatar
      Changes to toplevel configure[.in] to make the libntfs, and liblocal · 152a9958
      Kirk Webb authored
      subdirs and GNUmakefiles
      152a9958
    • Kirk Webb's avatar
      afabc26d
    • Kirk Webb's avatar
      Looks big and scary, and probably is. This checkin includes all the · 47962f80
      Kirk Webb authored
      cruft needed to compile ntfs support into imagezip, including the i18n
      unicode library
      routines from FBSD 5.0.  Most of the magic is in the makefiles, but there
      is a patch for the ntfsprogs package we're using to get NTFS support.
      (patch contributed by Russ).
      
      Note that the ntfsprogs tarball is located in emulab webspace (like TG).
      
      Don't worry!  You still have to pass --enable-windows to configure in order to
      pull any of this in, although it should compile nicely from here on out.
      
      Russ will commit his imagezip changes (hiber, and page file skipping) soon.
      47962f80
    • Leigh B. Stoller's avatar
      Tim Stack's long nits message. Lots of good ideas. Do not work on them · 8673563a
      Leigh B. Stoller authored
      without first checking with the power that be.
      8673563a
    • Robert Ricci's avatar
      Properly install an elvind config file, properly patched for use with · 4b5478e6
      Robert Ricci authored
      with the testbed.
      4b5478e6
    • Robert Ricci's avatar
      Two changes: · 4c25a593
      Robert Ricci authored
      First, don't select a plink for an emulated link that would cause us
      to go over bandwidth.
      
      Second, make FIX_PLINK_ENDPOINTS the default, and add a #define that
      makes it the default (instead of having to specify it on each plink
      line.)
      4c25a593
    • Chad Barb's avatar
      · 6a449d22
      Chad Barb authored
      tbswap: re-enabled retry (Undid leigh's last change)
      
      assign_wrapper.in:
         was left-joining reserved to nodes to get reserved list;
         This didn't get delays.
      
         now am doing separate query on reserved, and putting those
         into the %fixed and %alreadyAllocated hashes.
      6a449d22
    • Leigh B. Stoller's avatar
      Add link to spewlogfile for Mike. All the support was in the perl · 4e4adae4
      Leigh B. Stoller authored
      backend, but given that you have to mighty quick on the click, I never
      bothered to put in the link. But I hear Mike is the quickest clicker
      in the group.
      4e4adae4
    • Leigh B. Stoller's avatar
      Turn debugging off. · 4f9e6fdb
      Leigh B. Stoller authored
      4f9e6fdb
    • Leigh B. Stoller's avatar
      A possible hack/fix to the concurrency problem. I have not learned · 75936b39
      Leigh B. Stoller authored
      anything conclusive, other than fork confuses mysqld, and our attempt
      to unconfuse it by closing/opening the DB connection in the child,
      confuses it even more! The parent connection is shutdown and then
      restarted as well. In node_reboot, this could happen 10 times in rapid
      succession as we fork as child for each node, and that appears to
      strain the server even more (with 20 rapid connects).
      
      So, I added a select(0.3) in TBdbfork() (in the child) to see if that
      slows things down enough. In my tests, the length of time between the
      parent and child operations makes a real difference. With a delay in
      the child, the parent connection is not closed/restarted.
      
      The astute reader will realize that I have no actual clue; just
      grasping at, uh, threads.
      75936b39