[OLPC-devel] Re: [One Laptop Per Child] #14: Time synchronization

[Zarro Boogs per Child]

#14: Time synchronization
--------------------+-------------------------------------------------------
 Reporter:  jg      |        Owner:  ksankar   
     Type:  task    |       Status:  assigned  
 Priority:  normal  |    Milestone:  rev1 final
Component:  distro  |   Resolution:            
 Keywords:          |  
--------------------+-------------------------------------------------------
Changes (by krstic):

  * milestone:  rev1 beta => rev1 final

Old description:

> Clients
> -------
> For multiplayer mesh networks to work at all well for sound and games, we
> need very tight, robust time synchronization, even more fun when you
> think about our suspend/resume for power management problem.
>
> We don't yet know how accurate the TOY clocks may be on the machine,
> (typically not very), and the latency of a mesh network, particularly in
> areas requiring long hops and multiple hops, will be much more than is
> typically seen.
>
> Questions:
>   Will existing NTP daemons cope with this? (if any)
>   If not, how do we get something that does?
>   If yes, which one copes best with this environment?
>   How quickly can two machines that begin communications for applications
> needing tight synchronization that are out of sync
> become synchronized?  Do we need application level help?
>
> Servers
> -------
> Which NTP server should school servers run?
> Autoconfiguration of the time server.  We don't want manual
> intervention to be necessary.  Anycast to find upstream servers,
> maybe?
>
> Will schools want local time sources to use (e.g. GPS recievers).

New description:

 == Clients ==

 For multiplayer mesh networks to work at all well for sound and games, we
 need very tight, robust time synchronization, even more fun when you think
 about our suspend/resume for power management problem.

 We don't yet know how accurate the TOY clocks may be on the machine,
 (typically not very), and the latency of a mesh network, particularly in
 areas requiring long hops and multiple hops, will be much more than is
 typically seen.

 Questions:
  * Will existing NTP daemons cope with this? (if any)
  * If not, how do we get something that does?
  * If yes, which one copes best with this environment?
  * How quickly can two machines that begin communications for applications
 needing tight synchronization that are out of sync become synchronized?
 Do we need application level help?

 == Servers ==
  * Which NTP server should school servers run?
  * Autoconfiguration of the time server.  We don't want manual
 intervention to be necessary.  Anycast to find upstream servers, maybe?
  * Will schools want local time sources to use (e.g. GPS receivers).

Comment:

 No, permanent connectivity may certainly not be assumed.

-- 
Ticket URL: <http://dev.laptop.org/ticket/14#comment:4>
One Laptop Per Child <http://laptop.org/>