Re: v7.1b4 bad performance

FWIW, I get the following pgbench results on my machine (HPPA C180,
fast-wide-SCSI drives that I do not recall the specs for):

current sources, with -F

$ pgbench -t 1000 bench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 1
number of transactions per client: 1000
number of transactions actually processed: 1000/1000
tps = 26.493155(including connections establishing)
tps = 26.558319(excluding connections establishing)
$ pgbench -c 10 -t 100 bench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 10
number of transactions per client: 100
number of transactions actually processed: 1000/1000
tps = 25.812518(including connections establishing)
tps = 26.161266(excluding connections establishing)

current sources, without -F

$ pgbench -t 1000 bench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 1
number of transactions per client: 1000
number of transactions actually processed: 1000/1000
tps = 12.843274(including connections establishing)
tps = 12.864183(excluding connections establishing)
$ pgbench -c 10 -t 100 bench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 10
number of transactions per client: 100
number of transactions actually processed: 1000/1000
tps = 12.593353(including connections establishing)
tps = 12.676020(excluding connections establishing)

7.0.2, with -F

$ pgbench -p 5432 -t 1000 bench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 1
number of transactions per client: 1000
number of transactions actually processed: 1000/1000
tps = 48.925826(including connections establishing)
tps = 49.199684(excluding connections establishing)
$ pgbench -p 5432 -c 10 -t 100 bench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 10
number of transactions per client: 100
number of transactions actually processed: 1000/1000
tps = 43.664810(including connections establishing)
tps = 45.067229(excluding connections establishing)

7.0.2, without -F

$ pgbench -p 5432 -t 1000 bench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 1
number of transactions per client: 1000
number of transactions actually processed: 1000/1000
tps = 5.678665(including connections establishing)
tps = 5.682127(excluding connections establishing)
$ pgbench -p 5432 -c 10 -t 100 bench
starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 1
number of clients: 10
number of transactions per client: 100
number of transactions actually processed: 1000/1000
tps = 5.780491(including connections establishing)
tps = 5.796646(excluding connections establishing)

In short, about 2x faster when you compare the fsync (no -F) cases,
but slower when you compare the no-fsync cases. This may just be
because current sources have to do more I/O to write the WAL log as
well as the data files, but I'm not convinced of that... trying to
get some profile info ...

regards, tom lane