In a pureScale system with a high XA transaction workload, there
is a possible performance bottleneck in the existing SCA search
and insert logic. This APAR has been opened to address that
issue and devise a better workflow to help alleviate thie
potential bottleneck.
Formatting is a little messy, but from the CF monitoring you
will see a significant value for the request time element for
the ReadCastoutClass compared to the other calls:
=============================================================
CF-side command execution counts and average response times
=============================================================
select d.id, cast(substr(cf_cmd_name,1,50) as varchar(50)) as
cf_cmd_name, total_cf_requests, decimal( (total_cf_requests /
double(total_total_cf_requests)) * 100, 5, 2 ) as
pct_total_cf_cmd, decimal( double(total_cf_cmd_time_micro) /
total_cf_requests, 15, 2 ) as avg_cf_request_time_micro from
mon_get_cf_cmd_diff as d, ( select id, sum(total_cf_requests) as
total_total_cf_requests from mon_get_cf_cmd_diff group by id )
as c where d.total_cf_requests > 0 and d.id = c.id order by id
asc, total_cf_requests desc with UR
ID CF_CMD_NAME
TOTAL_CF_REQUESTS PCT_TOTAL_CF_CMD AVG_CF_REQUEST_TIME_MICRO
------ --------------------------------------------------
-------------------- ---------------- -------------------------
128 MessageResponseBlock
5945551 34.73 1.06
128 ProcessSetLockState
3824254 22.34 2.23
(...)
128 WriteAndRegisterMultiple
234652 1.37 99.02
128 ReadCastoutClass
222867 1.30 718.28 <==
128 CrossInvalidate
222353 1.29 8.99
128 ReadSetLFS
75930 0.44 1.70
128 ReadAndRegister
21980 0.12 3.14 |
