Miklos Szigetvari wrote:
To gather trace data from an application I can select GTF or CTRACE
to write external trace data
From the speed and overhead which is the better choise ?
Apologies to those of you seeing this twice. I responded first to my
local newsgroup feed.
I agree with Mark Jacobs in recommending CTRACE over GTF. GTF exploits
the monitor call instruction and the limited monitor call masking that
was designed in the mid-1960s. Regrettably, as the system has grown
more complex, the reliance made by GTF on the lowest-level serialization
technique used by z/OS, disablement, has made it less desirable than
letting components employ serialization as decentralized as possible
during tracing - the CTRACE approach.
Unauthorized CTRACE APIs allow you to start exploiting CTRACE by (1)
supplying a CTRACE statement in IPCS's parmlib input, (2) ensure that
trace tables get dumped when appropriate, (3) write a component find
routine, and (4) use trace entries that CTRACE's default formatting can
format tolerably. That's not a huge cost of entry, and subsequent
action can improve your results incrementally. The starting point of
in-storage tracing lets your component tolerate tracing that would be
excessive if an attempt were made to write entries to a data set, one
action that you might contemplate as an enhancement once you have
implemented your tracing and tuned the level to balance density with
intrusiveness.
Bob Wright - z/OS Service Aids
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
