Things that could be added:
    logging cleanup
    scheduler stats structures cleanup
    adminUp/adminDown host
    'weighted' hosts

    more doco
    write a client that compares config xml and running xml, and updates
      the latter to match the former. mostly there now, see compareconf
    Figure out a way to restart/upgrade pydir. Idea: close down listener, start
      new code immediately. Let old connections on old pydir finish (run to 
      completion) before exiting.
    Replace crappy manager thread and admin thread with twisted-based 
      implementations.

    Caching of DNS lookups
    dump out curBadHosts/histBadHosts/open
    make the connection checker work
    sticky connections. keep a cache of 'client_ip':'backend_ip'. 
      manager needs to be able to call "scheduler.cleanup()" to allow the
      cache to be manageable.

Admin interface bits left to do:

  Information:
    [maybe] config / config.txt ?
    [maybe] config vs. running
    [maybe] reload config, making changes
  Host mgmt:
    delAllHosts?service=NNN&group=NNN
  Group mgmt:
    changeScheduler?service=NNN&group=NNN&scheduler=NNN
  Service mgmt:
    newService?service=NNN&listen=NNN
    newGroup?service=NNN&group=NNN&scheduler=NNN
    delGroup?service=NNN&group=NNN
  other
    closeLog
    retryHost?service=NNN&group=NNN&client=NNN
    hostAdminDown
    hostAdminUp
    [maybe] shutdown?

multiple admins for a pydir (e.g. one http, one https)

make enable optional

logging options?

Document the various return results from the text mode interfaces - 
eg running.txt

additional schedulers.

better testing stuff in SchedulerManager, rather than just dropping the
dead entry back into the mix. In the case of "it's down", this is harmless,
but nonetheless. Store away more info about the down hosts - when, how long
til next test, &c.

expires on the CSS.

seriously refactor the CSS and also the pdadmin_running HTML. Nasty nasty.

pid file?

specify users not in the config.xml file, but somewhere else.

what does 'host unknown' raise under 2.1? 

Clean up all the logging. Logging now goes to a single place, but in
inconsistent format. Still, it's better than it was.

Track the count of faults.

If M2Crypto is installed, allow a https-based admin interface. Sort-of
there now, use the undocumented 'secure="yes"' flag to the admin directive.

Additional idea: connection count scheduler. could be used to track 
licensing of a service, or limit it.

