RE: Delaying/avoiding BTreeTupleGetNAtts() call within _bt_compare() - Mailing list pgsql-hackers

> 
> Can you test this version, Floris? The second two patches are probably not
> helping here, so it would be useful if you could just test 0001-*, and then test
> all three together. I can toss the latter two patches if there is no additional
> speedup.
> 

Here's the results for runs with respectively 1 client, 9 clients and 30 clients on master, v2-0001, v2-0001+0002+0003
andfor completeness also the previous v1 version of the patches.
 
I ran the tests for 45 minutes each this time which seems to give more stable results.
I'd say applying just v2-0001 is actually slightly hurtful for single-core performance. Applying all of them gives a
goodimprovement though. It looks like the performance improvement is also more noticeable at higher core counts now.
 

<master>
number of clients: 1
number of threads: 1
duration: 2700 s
number of transactions actually processed: 139314796
latency average = 0.019 ms
latency stddev = 0.001 ms
tps = 51598.071835 (including connections establishing)
tps = 51598.098715 (excluding connections establishing)

<v2-0001>
number of clients: 1
number of threads: 1
duration: 2700 s
number of transactions actually processed: 137257492
latency average = 0.020 ms
latency stddev = 0.001 ms
tps = 50836.107076 (including connections establishing)
tps = 50836.133137 (excluding connections establishing)

<v2-0001+2+3>
number of clients: 1
number of threads: 1
duration: 2700 s
number of transactions actually processed: 141721881
latency average = 0.019 ms
latency stddev = 0.001 ms
tps = 52489.584928 (including connections establishing)
tps = 52489.611373 (excluding connections establishing)

<v1-0001+2+3>
number of clients: 1
number of threads: 1
duration: 2700 s
number of transactions actually processed: 141663780
latency average = 0.019 ms
latency stddev = 0.001 ms
tps = 52468.065549 (including connections establishing)
tps = 52468.093018 (excluding connections establishing)



<master>
number of clients: 9
number of threads: 9
duration: 2700 s
number of transactions actually processed: 1197242115
latency average = 0.020 ms
latency stddev = 0.001 ms
tps = 443422.987601 (including connections establishing)
tps = 443423.306495 (excluding connections establishing)

<v2-0001>
number of clients: 9
number of threads: 9
duration: 2700 s
number of transactions actually processed: 1187890004
latency average = 0.020 ms
latency stddev = 0.002 ms
tps = 439959.241392 (including connections establishing)
tps = 439959.588125 (excluding connections establishing)

<v2-0001+2+3>
number of clients: 9
number of threads: 9
duration: 2700 s
number of transactions actually processed: 1203412941
latency average = 0.020 ms
latency stddev = 0.002 ms
tps = 445708.478915 (including connections establishing)
tps = 445708.798583 (excluding connections establishing)

<v1-0001+2+3>
number of clients: 9
number of threads: 9
duration: 2700 s
number of transactions actually processed: 1195359533
latency average = 0.020 ms
latency stddev = 0.001 ms
tps = 442725.734267 (including connections establishing)
tps = 442726.052676 (excluding connections establishing)


<master>
number of clients: 30
number of threads: 30
duration: 2700 s
number of transactions actually processed: 2617037081
latency average = 0.031 ms
latency stddev = 0.011 ms
tps = 969272.811990 (including connections establishing)
tps = 969273.960316 (excluding connections establishing)

<v2-0001>
number of clients: 30
number of threads: 30
duration: 2700 s
number of transactions actually processed: 2736881585
latency average = 0.029 ms
latency stddev = 0.011 ms
tps = 1013659.581348 (including connections establishing)
tps = 1013660.819277 (excluding connections establishing)

<v2-0001+2+3>
number of clients: 30
number of threads: 30
duration: 2700 s
number of transactions actually processed: 2844199686
latency average = 0.028 ms
latency stddev = 0.011 ms
tps = 1053407.074721 (including connections establishing)
tps = 1053408.220093 (excluding connections establishing)

<v1-0001+2+3>
number of clients: 30
number of threads: 30
duration: 2700 s
number of transactions actually processed: 2693765822
latency average = 0.030 ms
latency stddev = 0.011 ms
tps = 997690.883117 (including connections establishing)
tps = 997692.051005 (excluding connections establishing)