[Beowulf] IEEE 1588 (PTP) - a better cluster clock?

[Jim Lux]

At 01:11 PM 7/24/2007, Patrick Geoffray wrote:
>Hi Jim,
>
>Jim Lux wrote:
>>Highly parallelized real time signal processing?  Seems like a classic
>
>Wouldn't you need a real-time OS and a real-time communication layer 
>to do real-time processing? Or at least within the same level of 
>time accuracy ?

not necessarily.

>The Linux scheduler is still on a 10ms quantum of time, and QoS in 
>network is still about what to drop first, not guaranteeing resources.

Consider something where you had something like a frame grabber. you 
might need microsecond precision for some aspects, but just 
guarantees of sufficient resources for farther back in the chain.

>If PTP is as good (accuracy, resource usage) as the specialized 
>methods, then using a common implementation is the best deal. 
>However, application-specific/architecture-specific synchronization 
>may be more efficient. The trade-off is between the performance of 
>the first versus the cost of the second. That's the rule behind 
>commodity solutions.

Indeed..

>Patrick

James Lux, P.E.
Spacecraft Radio Frequency Subsystems Group
Flight Communications Systems Section
Jet Propulsion Laboratory, Mail Stop 161-213
4800 Oak Grove Drive
Pasadena CA 91109
tel: (818)354-2075
fax: (818)393-6875