On Mon, 20 Jun 2022 21:55:07 -0700, Charles Mills wrote:
>> But is loss of a single record disastrous?
>
>No, I said earlier, this is not banking transactions where a lack of 2-phase
>commit risks losing a bunch of money between the cracks.
>
What's your crash recovery plan?
o Create a new file and process both at report time?
o Continue writing to the same file with O_APPEND?
Long ago, I reported an apparent I/O performance problem. Support examined
my code and explained that I used O_APPEND, which incurs significant
serialization overhead. I assume this is to preserve integrity if multiple jobs
write to the same log file.
(I mentioned this on MVS-OE and WJS quickly jumped in, saying he didn't
recognize the behavior, but a sysplex might add overhead. He'll be missed
when he retires at the end of the month.)
The cost of O_APPEND is likely to be insignificant for your expected traffic.
And you don't expect multiple writers. I'd expect it to be worse for Classic
data sets:
ALLOCATE DISP=MOD
OPEN
PUT one record
CLOSE
FREE
(Adverse blocking!)
--
gil
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
