As the author of some SNMP code, I'd like to see the *whole*
of /proc adopt a consistent, simple to parse, simple to read,
simple to write structure.
Also needed is a (user space) means for accessing the meta-data
for /proc data values. For example, a number means nothing without
knowning if it is an accumulator or the current value.
The lack of meta-data is the reason we hear so many "what does
this variable mean?" questions.
It would then be realistic to port tools for the tuning of large
Linux clusters.
At the moment, the configuration of performance visualisation
tools has to be done by hand for every variable. This defeats
the whole point of visualisation and exploratory data analysis.
>From the API side, it has to be simpler for drivers to maintain
counters. The current level of kernel instrumentation really
sucks when compared to other unix-like OSs.
In terms of breaking user-space apps, I think most user-space
authors would prefer to see one big "let's get it right" change,
rather than the current situation where formats can change
radically with device driver sub-versions and the user-space
application author is always at one step behind for at least some
of the operating system.
-- Glen Turner Network Specialist Tel: (08) 8303 3936 Information Technology Services Fax: (08) 8303 4400 The University of Adelaide 5005 Email: glen.turner@adelaide.edu.au South Australia- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/