perfsonar development work

[Jason Zurawski <>]

All;

>>>>
>>>> I'd like to ask about the interval time for LS registering. So far
>>>> it's 30 seconds for RRD MA and SQL MA. Do you want to keep this value
>>>> or increase for the release 3.0 (personally I think it should be
>>>> higher value, 10 mins or more; currently LS removes service's
>>>> metadatas if they are not sent/updated in new register request during
>>>> the last 24 hours)?
>>>>       
>>> We prefer it to be waaaay higher than 30 seconds. At least a few hours.
>>>
>>> But we don't have good analysis of the impact to backup our
>>> suggestions.
>>>
>>>    
>>>> Roman
>>>>       
>>
>> Isn't this an administrative issue? The systemn administrator who
>> installs the service can configure the interval time to whatever
>> value seems appropriate. For the BWCTL MP we have a value of 3600
>> seconds (one hour) as default in the configuration file.
>>   
> Many administrators don't change the defaults unless they have good
> reasons to do so. Our responsibility is to ensure that we choose very
> sensible default values.

Sort of offtopic...

Future iterations of the LS need to be able to report when they are
being throttled so that the registration interval of a single rogue
service does not harm the rest of the framework.  We are already
implementing steps that allows an LS to report back to registering
services the internal data TTL; this allows intelligent services the
ability to interpret this value and adjust their own registration
interval if necessary.

Currently the LS doesnt 'reject' messages if they are arriving too fast
(opening the messages to check who its from, when the last time
registered/next expected registration would still consume a lot of
resources), and we foresee this step as not being necessary once
resource protection makes an appearance.

-jason