[ntp:questions] Re: Advice on sync clock between cluster of linux v2.6 to +-1us

Helmut Wollmersdorfer helmut at wollmersdorfer.at
Wed Jan 12 11:41:32 UTC 2005


Nick wrote:

> I've spent the morning doing some research on synchronising a cluster of 
> computers running linux. 

Which sort of cluster?

> I don't require the computers to have the correct actual time, only the 
> time between them needs to be accurate.

Why?
It's not a good idea to use software for critical applications which is 
time sensitive in a way, that it would crash, or produce wrong results.

E.g. for financial applications like ledger or telephone billing it can 
be an important requirement, that system clocks can be apart months or 
years. Testing if the application logic does not depend on system clocks 
is sometimes called "time travelling". On the other hand interfaces to 
the real word need real time in _production_ to guarantee accurate logs, 
traceability etc.

On my heartbeat+DRBD cluster, DRBD is time insensitive, as it uses 
generation counts instead of time stamps across the nodes. Heartbeat is 
time sensitive in terms of minutes (or larger values of seconds).

> I don't believe I need (the expense of) a GPS/Radio receiver since I 
> don't require the actual time, but a lot of these receivers have a TCXO 
> (or alternative) module to handle failover due to signal failure.

Referencing a public time server costs nearly nothing and you can expect 
an accuracy of 10-20ms to real time and some ms between your nodes.

A cheap DIY (Do It Yourself) radio receiver like 
http://www.buzzard.org.uk/jonathan/radioclock.html costs around
USD 50, some hours work, and has accuracy of a few ms to real time.

Going to better accuracy costs more time and money, with GPS at USD 
200-500 providing the best cost/benefit ratio.

Helmut Wollmersdorfer



More information about the questions mailing list