On Thu, Nov 13, 2014 at 7:03 PM, Heikki Linnakangas <hlinnakan...@vmware.com> wrote: > > On 11/11/2014 04:42 PM, Amit Kapila wrote: >> >> I have done some performance testing of this patch using attached >> script and data is as below: >> >> ... >> >> It seems to me that there is a regression of (4 ~ 8%) for small records, >> refer two short fields tests. > > > Thanks for the testing! > > > Thoughts on this new WAL record format? I've attached the xlogrecord.h file here separately for easy reading, if you want to take a quick look at just that without applying the whole patch. >
Apart from changes in XLogRecord to remove xl_len and padding, the new format for block headers seems to be quite succinct and then by making data length as variable for actual record the overall WAL record size becomes smaller. However the increase in unaligned size (as mentined by you up-thread as 2 bytes) seems to be still remain same. Is the difference in size is due to additional block reference id this patch requires or something else? In new record format the Record header always start at aligned boundary, however the actual record data doesn't necessarily be at aligned boundary, can this make difference as currently we copy both of these separately? Overall, I think this format is a net improvement. With Regards, Amit Kapila. EnterpriseDB: http://www.enterprisedb.com