Poor dump performance on new machine

[Guillaume Couvez]

Hello,

We are experiencing a very strange decrease of dump performance on our
new hardware.

We are dumping a 67 Go database.

On the old hardware it takes 21 mins

AIX 5.1
IBM Nighthawk2 - 16 proc - 375MHz -16Go RAM
Dump location is a filesystem to 4 SSA disk 10.000 rpm (stripping)


On the new hardware it takes around 60 mins

AIX 5.3
IBM P570 - 4 proc - 1.65GHz - 16 Go RAM
Dump location is a filesystem to Fast600 - 128 Mo cache - disk bay IBM
(Raid 1+0 on 6 disk 10.000 rpm)

iostat on the new server shows IO waits when the backup server is
running at 20 Mb / sec. Unix Admin can run test at more than 80 Mb /
sec

We have the same configuration of Backup Server on both platform.
Anyone have an idea of what can be the cause of the poor performance?

Regards.

[unknown]

Post by Guillaume Couvez
Hello,
We are experiencing a very strange decrease of dump
performance on our new hardware.
We are dumping a 67 Go database.
On the old hardware it takes 21 mins
AIX 5.1
IBM Nighthawk2 - 16 proc - 375MHz -16Go RAM
Dump location is a filesystem to 4 SSA disk 10.000 rpm
(stripping)
On the new hardware it takes around 60 mins
AIX 5.3
IBM P570 - 4 proc - 1.65GHz - 16 Go RAM
Dump location is a filesystem to Fast600 - 128 Mo cache -
disk bay IBM (Raid 1+0 on 6 disk 10.000 rpm)
iostat on the new server shows IO waits when the backup
server is running at 20 Mb / sec. Unix Admin can run test
at more than 80 Mb / sec
We have the same configuration of Backup Server on both
platform. Anyone have an idea of what can be the cause of
the poor performance?
Regards.

How many engines you have configured in the new environment
you should leave 1 cpu free for OS processes.

[Jason L. Froebe (Team Sybase)]

Post by unknown

Post by Guillaume Couvez
Hello,
We are experiencing a very strange decrease of dump
performance on our new hardware.
We are dumping a 67 Go database.
On the old hardware it takes 21 mins
AIX 5.1
IBM Nighthawk2 - 16 proc - 375MHz -16Go RAM
Dump location is a filesystem to 4 SSA disk 10.000 rpm
(stripping)
On the new hardware it takes around 60 mins
AIX 5.3
IBM P570 - 4 proc - 1.65GHz - 16 Go RAM
Dump location is a filesystem to Fast600 - 128 Mo cache -
disk bay IBM (Raid 1+0 on 6 disk 10.000 rpm)
iostat on the new server shows IO waits when the backup
server is running at 20 Mb / sec. Unix Admin can run test
at more than 80 Mb / sec
We have the same configuration of Backup Server on both
platform. Anyone have an idea of what can be the cause of
the poor performance?
Regards.

How many engines you have configured in the new environment
you should leave 1 cpu free for OS processes.

A decade ago, that would be a valid statement but it is no longer true.

jason

[Juan C. Lopez]

We have an experience with that I think on february on hp-unix with 2
processors and 2 engines configured. Performance like crap until we set
the engines back to 1.

Post by Jason L. Froebe (Team Sybase)

Post by unknown

Post by Guillaume Couvez
Hello,
We are experiencing a very strange decrease of dump
performance on our new hardware.
We are dumping a 67 Go database.
On the old hardware it takes 21 mins
AIX 5.1
IBM Nighthawk2 - 16 proc - 375MHz -16Go RAM
Dump location is a filesystem to 4 SSA disk 10.000 rpm
(stripping)
On the new hardware it takes around 60 mins
AIX 5.3
IBM P570 - 4 proc - 1.65GHz - 16 Go RAM
Dump location is a filesystem to Fast600 - 128 Mo cache -
disk bay IBM (Raid 1+0 on 6 disk 10.000 rpm)
iostat on the new server shows IO waits when the backup
server is running at 20 Mb / sec. Unix Admin can run test
at more than 80 Mb / sec
We have the same configuration of Backup Server on both
platform. Anyone have an idea of what can be the cause of
the poor performance?
Regards.

How many engines you have configured in the new environment
you should leave 1 cpu free for OS processes.

A decade ago, that would be a valid statement but it is no longer true.
jason
--
Jason L. Froebe
http://jfroebe.livejournal.com
Team Sybase

[Jason L. Froebe (Team Sybase)]

Post by Juan C. Lopez
We have an experience with that I think on february on hp-unix with 2
processors and 2 engines configured. Performance like crap until we set
the engines back to 1.

Was it a dedicated box or were you running other applications?
Application server, etc. The os operations itself doesn't really
require much cpu time. If there was no other operations going on on the
hpux box, I would wonder about the configuration of the box, whether it
was inadvertently misconfigured or some other unidentified operation was
going on.

[Stefan Karlsson]

Post by unknown

Post by Guillaume Couvez
Hello,
We are experiencing a very strange decrease of dump
performance on our new hardware.
We are dumping a 67 Go database.
On the old hardware it takes 21 mins
AIX 5.1
IBM Nighthawk2 - 16 proc - 375MHz -16Go RAM
Dump location is a filesystem to 4 SSA disk 10.000 rpm
(stripping)
On the new hardware it takes around 60 mins
AIX 5.3
IBM P570 - 4 proc - 1.65GHz - 16 Go RAM
Dump location is a filesystem to Fast600 - 128 Mo cache -
disk bay IBM (Raid 1+0 on 6 disk 10.000 rpm)
iostat on the new server shows IO waits when the backup
server is running at 20 Mb / sec. Unix Admin can run test
at more than 80 Mb / sec
We have the same configuration of Backup Server on both
platform. Anyone have an idea of what can be the cause of
the poor performance?
Regards.

How many engines you have configured in the new environment
you should leave 1 cpu free for OS processes.

Have you checked vmstat ? I suspect that the host is paging. Perhaps the
AIX paper I co-wrote with IBM may help with vm configuration:
http://www-03.ibm.com/servers/enable/site/peducation/wp/b78a/index.html

HTH,

/Stefan