Re: pgbench Windows outperforms Debian in Write tests while Debian is faster in SELECT only tests

> -----Ursprüngliche Nachricht-----

> Von: Tom Lane [mailto:tgl(at)sss(dot)pgh(dot)pa(dot)us]

> Gesendet: Montag, 29. Mai 2017 10:19

> An: Klaus P. Pieper

> Cc: pgsql-admin(at)postgresql(dot)org

> Betreff: Re: [ADMIN] pgbench Windows outperforms Debian in Write tests

> while Debian is faster in SELECT only tests

>

> "Klaus P. Pieper" < <mailto:kpi6288(at)gmail(dot)com> kpi6288(at)gmail(dot)com> writes:

> > How can Windows outperform Debian by a factor of 2?

>

> Are you sure you have Windows configured to fsync properly?

> Not waiting for writes to hit disk would be a really easy explanation for
a

> discrepancy of this kind.

It seems that it is indeed related to the wal_sync_method settings.

Results for 1 client read / write:

Option Windows Debian

fdatasync 340 150

fsync 125 90

open_datasync 360 160

fsync_writethrough 115 n/a

open_sync n/a 90

fsync = off 360 400

The result setting fsync=off shows that the maximum performance of the
Debian syste is above Windows.

However, with fsync=on, Windows still outperforms the Debian system with all
options.

In addition, the results on the Windows system with “fsync = off” are the
same as “wal_sync_method = open_datasync”

What does this mean?

- Is the Windows system “better” than the Debian system?

- Or is the underlying filesystem on the Windows system less safe?

- Or is the Windows system better integrated with the Hyper-V host?

The Windows systems reports that the write cache is switched on for the
virtual disk. It is however not possible to switch is off.

Thanks

Klaus