perfsonar development work

[Nina Jeliazkova <>]

Hi Martin, All,

Martin Swany написа:
> Hi Nina,
>
>> You might remember (or not), earlier versions of PerfsonarUI were
>> ignoring the valueunit string. And everything was working fine, when
>> one service started to report values in bits rather than presumed
>> octets and PerfsonarUI was reporting utilization of 400%.
>>
>> Even if we consider only counters, the client should have the way of
>> recognizing that "errors/sec" and "Eps" are the same thing, in order
>> to plot them on the same axis, or report an average for the last
>> hour, for example. I believe it is not correct to plot apparently
>> different entities to the same axis or averaging between different
>> entities.
>> While for the human reader the semantics of both string is obviously
>> the same, for the client those are just strings, but DIFFERENT strings.
>>
>> My point is that we need a standardized way of valueunit string
>> interpretation, and the best thing coming to my mind is to have
>> standardized list of units. Any other suggestions are welcome.
>
> Very good points.  When there are percentages to be computed,
> or when we need to evaluate equivalency, the units are critical.
> I recall that we discussed "Eps", "errors/sec" and the like and
> made some decisions.  I don't recall those decisions.  If someone
> can remind me/point me in the right direction, I'll include them in
> the protocol document draft in preparation.
Well, the discussion about value units had been on the list several
times, but nonetheless, no standardized list appeared so far.

Regarding errors and drops, perfsonarUI implementation is based on
examples Jason sent me in the beginning of August and were later
implemented by MA services.  With interpretation of units in perfsonarUI
being essential,  and no other source of information available, we took
a restrictive approach of allowing only "Eps" and "Dps".

Finally, I hope this time we'll come to a conclusion the standardization
is a good thing, especially when XML responses are expected to be
interpreted by a software and not only by human eyes.

A similar issue seems to appear with ifAddress/ipAddress representing
the same entity (IP address), as noticed by Fausto. 

Best regards,
Nina
>
> thanks,,
> martin
>