[vwnc] A question on PC clock time ...

> I have a client with a couple of dell duo-cpu workstations
> which tend to freeze in VW for seconds at a time (10 seconds, 30 seconds).
> This happens at least a few times an hour maybe more.
>
> We had a server which exhibited worse symptoms than that which was
> running Windows on XEN with multiple cpu's -- seemed that when
> a different cpu was selected the clock would be out a bit and that sent
> Delay's into snits -- we fixed that with "affinity" to a single cpu.
>
> Since that was a clock problem, I decided to watch the clocks on the
> offending workstations.  I can watch the clock gain a couple of seconds
> in a minute and then get "fixed" by the domain server presumably.  I have
> seen it off by as much as 8 seconds before getting "fixed".
>
> Has anyone seen problems with VW with this type of thing??
>
>  

> Dennis Smith wrote:
>> I have a client with a couple of dell duo-cpu workstations
>> which tend to freeze in VW for seconds at a time (10 seconds, 30
>> seconds).
>> This happens at least a few times an hour maybe more.
>>
>> We had a server which exhibited worse symptoms than that which was
>> running Windows on XEN with multiple cpu's -- seemed that when
>> a different cpu was selected the clock would be out a bit and that sent
>> Delay's into snits -- we fixed that with "affinity" to a single cpu.
>>
>> Since that was a clock problem, I decided to watch the clocks on the
>> offending workstations.  I can watch the clock gain a couple of seconds
>> in a minute and then get "fixed" by the domain server presumably.  I
>> have
>> seen it off by as much as 8 seconds before getting "fixed".
>>
>> Has anyone seen problems with VW with this type of thing??
>>
>>  
> Yep. It can be a real problem ,typically on Windows more than Linux or
> other Unixes. Martin and I have been working on a new Timer class that
> doesn't suffer from this kind of problem - it uses more modern timer
> APIs that are aware of the way the clock can shift with NTP. I don't
> think multicore exacerbates the problem though.

>
>
> Michael Lucas-Smith wrote:
>> Dennis Smith wrote:
>>> I have a client with a couple of dell duo-cpu workstations
>>> which tend to freeze in VW for seconds at a time (10 seconds, 30
>>> seconds).
>>> This happens at least a few times an hour maybe more.
>>>
>>> We had a server which exhibited worse symptoms than that which was
>>> running Windows on XEN with multiple cpu's -- seemed that when
>>> a different cpu was selected the clock would be out a bit and that sent
>>> Delay's into snits -- we fixed that with "affinity" to a single cpu.
>>>
>>> Since that was a clock problem, I decided to watch the clocks on the
>>> offending workstations.  I can watch the clock gain a couple of seconds
>>> in a minute and then get "fixed" by the domain server presumably.  I
>>> have
>>> seen it off by as much as 8 seconds before getting "fixed".
>>>
>>> Has anyone seen problems with VW with this type of thing??
>>>
>>>  
>> Yep. It can be a real problem ,typically on Windows more than Linux
>> or other Unixes. Martin and I have been working on a new Timer class
>> that doesn't suffer from this kind of problem - it uses more modern
>> timer APIs that are aware of the way the clock can shift with NTP. I
>> don't think multicore exacerbates the problem though.
> Will that be available as a patch??