hardware upgrade, performance degrade?

[Steven Crandell]

Recently I moved my ~600G / ~15K TPS database from a
48 ***@2.0GHz server with 512GB RAM on 15K RPM disk
to a newer server with
64 ***@2.2Ghz server with 1T of RAM on 15K RPM disks

The move was from v9.1.4 to v9.1.8 (eventually also tested with v9.1.4 on
the new hardware) and was done via base backup followed by slave promotion.
All postgres configurations were matched exactly as were system and kernel
parameters.

On the first day that this server saw production load levels it absolutely
fell on its face. We ran an exhaustive battery of tests including failing
over to the new (hardware matched) slave only to find the problem happening
there also.

After several engineers all confirmed that every postgres and system
setting matched, we eventually migrated back onto the original hardware
using exactly the same methods and settings that had been used while the
data was on the new hardware. As soon as we brought the DB live on the
older (supposedly slower) hardware, everything started running smoothly
again.

As far as we were able to gather in the frantic moments of downtime,
hundreds of queries were hanging up while trying to COMMIT. This in turn
caused new queries backup as they waited for locks and so on.

Prior to failing back to the original hardware, we found interesting posts
about people having problems similar to ours due to NUMA and several
suggested that they had solved their problem by setting
vm.zone_reclaim_mode = 0

Unfortunately we experienced the exact same problems even after turning off
the zone_reclaim_mode. We did extensive testing of the i/o on the new
hardware (both data and log arrays) before it was put into service and
have done even more comprehensive testing since it came out of service.
The short version is that the disks on the new hardware are faster than
disks on the old server. In one test run we even set the server to write
WALs to shared memory instead of to the log LV just to help rule out i/o
problems and only saw a marginal improvement in overall TPS numbers.

At this point we are extremely confident that if we have a configuration
problem, it is not with any of the usual postgresql.conf/sysctl.conf
suspects. We are pretty sure that the problem is being caused by the
hardware in some way but that it is not the result of a hardware failure
(e.g. degraded array, raid card self tests or what have you).

Given that we're dealing with new hardware and the fact that this still
acts a lot like a NUMA issue, are there other settings we should be
adjusting to deal with possible performance problems associated with NUMA?

Does this sound like something else entirely?

Any thoughts appreciated.

thanks,
Steve

[Guillaume Smet]

Hi Steven,

On Fri, Mar 1, 2013 at 10:52 AM, Steven Crandell

Given that we're dealing with new hardware and the fact that this still acts
a lot like a NUMA issue, are there other settings we should be adjusting to
deal with possible performance problems associated with NUMA?

On my kernel related thread list, I have:
http://www.postgresql.org/message-id/CAOR=d=2tjWoxpQrUHpJK1R+BtEvBv4buiVtX4Qf6we=***@mail.gmail.com
(you haven't mentioned if the kernel version is identical for both
servers)
http://www.postgresql.org/message-id/***@optionshouse.com
http://www.postgresql.org/message-id/***@optionshouse.com
http://www.postgresql.org/message-id/CAMAsR=5F45+kj+hw9q+zE7zo=Qc0yBEB1sLXCF0QL+***@mail.gmail.com

Might be of some interest to you.

[Craig James]

On Fri, Mar 1, 2013 at 1:52 AM, Steven Crandell

Post by Steven Crandell
Recently I moved my ~600G / ~15K TPS database from a
to a newer server with
The move was from v9.1.4 to v9.1.8 (eventually also tested with v9.1.4 on
the new hardware) and was done via base backup followed by slave promotion.
All postgres configurations were matched exactly as were system and kernel
parameters.
On the first day that this server saw production load levels it absolutely
fell on its face. We ran an exhaustive battery of tests including failing
over to the new (hardware matched) slave only to find the problem happening
there also.
After several engineers all confirmed that every postgres and system
setting matched, we eventually migrated back onto the original hardware
using exactly the same methods and settings that had been used while the
data was on the new hardware. As soon as we brought the DB live on the
older (supposedly slower) hardware, everything started running smoothly
again.
As far as we were able to gather in the frantic moments of downtime,
hundreds of queries were hanging up while trying to COMMIT. This in turn
caused new queries backup as they waited for locks and so on.
Prior to failing back to the original hardware, we found interesting posts
about people having problems similar to ours due to NUMA and several
suggested that they had solved their problem by setting
vm.zone_reclaim_mode = 0
Unfortunately we experienced the exact same problems even after turning
off the zone_reclaim_mode. We did extensive testing of the i/o on the new
hardware (both data and log arrays) before it was put into service and
have done even more comprehensive testing since it came out of service.
The short version is that the disks on the new hardware are faster than
disks on the old server. In one test run we even set the server to write
WALs to shared memory instead of to the log LV just to help rule out i/o
problems and only saw a marginal improvement in overall TPS numbers.
At this point we are extremely confident that if we have a configuration
problem, it is not with any of the usual postgresql.conf/sysctl.conf
suspects. We are pretty sure that the problem is being caused by the
hardware in some way but that it is not the result of a hardware failure
(e.g. degraded array, raid card self tests or what have you).
Given that we're dealing with new hardware and the fact that this still
acts a lot like a NUMA issue, are there other settings we should be
adjusting to deal with possible performance problems associated with NUMA?
Does this sound like something else entirely?
Any thoughts appreciated.

One piece of information that you didn't supply ... sorry if this is
obvious, but did you run the usual range of performance tests using
pgbench, bonnie++ and so forth to confirm that the new server was working
well before you put it into production? Did it compare well on those same
tests to your old hardware?

Craig

Post by Steven Crandell
thanks,
Steve

[Steven Crandell]

We saw the same performance problems when this new hardware was running
cent 6.3 with a 2.6.32-279.19.1.el6.x86_64 kernel and when it was matched
to the OS/kernel of the old hardware which was cent 5.8 with
a 2.6.18-308.11.1.el5 kernel.

Yes the new hardware was thoroughly tested with bonnie before being put
into services and has been tested since. We are unable to find any
interesting differences in our bonnie tests comparisons between the old and
new hardware. pgbench was not used prior to our discovery of the problem
but has been used extensively since. FWIW This server ran a zabbix
database (much lower load requirements) for a month without any problems
prior to taking over as our primary production DB server.

After quite a bit of trial and error we were able to find a pgbench test
(2x 300 concurrent client sessions doing selects along with 1x 50
concurrent user session doing the standard pgbench query rotation) that
showed the new hardware under performing when compared to the old hardware
to the tune of about a 1000 TPS difference (2300 to 1300) for the 50
concurrent user pgbench run and about a 1000 less TPS for each of the
select only runs (~24000 to ~23000). Less demanding tests would be handled
equally well by both old and new servers. More demanding tests would tip
both old and new over with very similar efficacy.

Hopefully that fleshes things out a bit more.
Please let me know if I can provide additional information.

thanks
steve

Post by Craig James

Post by Steven Crandell
Recently I moved my ~600G / ~15K TPS database from a
to a newer server with
The move was from v9.1.4 to v9.1.8 (eventually also tested with v9.1.4 on
the new hardware) and was done via base backup followed by slave promotion.
All postgres configurations were matched exactly as were system and
kernel parameters.
On the first day that this server saw production load levels it
absolutely fell on its face. We ran an exhaustive battery of tests
including failing over to the new (hardware matched) slave only to find the
problem happening there also.
After several engineers all confirmed that every postgres and system
setting matched, we eventually migrated back onto the original hardware
using exactly the same methods and settings that had been used while the
data was on the new hardware. As soon as we brought the DB live on the
older (supposedly slower) hardware, everything started running smoothly
again.
As far as we were able to gather in the frantic moments of downtime,
hundreds of queries were hanging up while trying to COMMIT. This in turn
caused new queries backup as they waited for locks and so on.
Prior to failing back to the original hardware, we found interesting
posts about people having problems similar to ours due to NUMA and several
suggested that they had solved their problem by setting
vm.zone_reclaim_mode = 0
Unfortunately we experienced the exact same problems even after turning
off the zone_reclaim_mode. We did extensive testing of the i/o on the new
hardware (both data and log arrays) before it was put into service and
have done even more comprehensive testing since it came out of service.
The short version is that the disks on the new hardware are faster than
disks on the old server. In one test run we even set the server to write
WALs to shared memory instead of to the log LV just to help rule out i/o
problems and only saw a marginal improvement in overall TPS numbers.
At this point we are extremely confident that if we have a configuration
problem, it is not with any of the usual postgresql.conf/sysctl.conf
suspects. We are pretty sure that the problem is being caused by the
hardware in some way but that it is not the result of a hardware failure
(e.g. degraded array, raid card self tests or what have you).
Given that we're dealing with new hardware and the fact that this still
acts a lot like a NUMA issue, are there other settings we should be
adjusting to deal with possible performance problems associated with NUMA?
Does this sound like something else entirely?
Any thoughts appreciated.

One piece of information that you didn't supply ... sorry if this is
obvious, but did you run the usual range of performance tests using
pgbench, bonnie++ and so forth to confirm that the new server was working
well before you put it into production? Did it compare well on those same
tests to your old hardware?
Craig

Post by Steven Crandell
thanks,
Steve

[Scott Marlowe]

On Fri, Mar 1, 2013 at 9:49 AM, Steven Crandell

We saw the same performance problems when this new hardware was running cent
6.3 with a 2.6.32-279.19.1.el6.x86_64 kernel and when it was matched to the
OS/kernel of the old hardware which was cent 5.8 with a 2.6.18-308.11.1.el5
kernel.
Yes the new hardware was thoroughly tested with bonnie before being put into
services and has been tested since. We are unable to find any interesting
differences in our bonnie tests comparisons between the old and new
hardware. pgbench was not used prior to our discovery of the problem but
has been used extensively since. FWIW This server ran a zabbix database
(much lower load requirements) for a month without any problems prior to
taking over as our primary production DB server.
After quite a bit of trial and error we were able to find a pgbench test (2x
300 concurrent client sessions doing selects along with 1x 50 concurrent
user session doing the standard pgbench query rotation) that showed the new
hardware under performing when compared to the old hardware to the tune of
about a 1000 TPS difference (2300 to 1300) for the 50 concurrent user
pgbench run and about a 1000 less TPS for each of the select only runs
(~24000 to ~23000). Less demanding tests would be handled equally well by
both old and new servers. More demanding tests would tip both old and new
over with very similar efficacy.
Hopefully that fleshes things out a bit more.
Please let me know if I can provide additional information.

OK I'd recommend testing with various numbers of clients and seeing
what kind of shape you get from the curve when you plot it. I.e. does
it fall off really hard at some number etc? If the old server
degrades more gracefully under very heavy load it may be that you're
just admitting too many connections for the new one etc, not hitting
its sweet spot.

FWIW, the newest intel 10 core xeons and their cousins just barely
keep up with or beat the 8 or 12 core AMD Opterons from 3 years ago in
most of my testing. They look great on paper, but under heavy load
they are luck to keep up most the time.

There's also the possibility that even though you've turned off zone
reclaim that your new hardware is still running in a numa mode that
makes internode communication much more expensive and that's costing
you money. This may especially be true with 1TB of memory that it's
both running at a lower speed AND internode connection costs are much
higher. use the numactl command (I think that's it) to see what the
internode costs are, and compare it to the old hardware. IF the
internode comm costs are really high, see if you can turn off numa in
the BIOS and if it gets somewhat better.

Of course check the usual, that your battery backed cache is really
working in write back not write through etc.

Good luck. Acceptance testing can really suck when newer, supposedly
faster hardware is in fact slower.

[Josh Berkus]

Steven,

Post by Steven Crandell
We saw the same performance problems when this new hardware was running
cent 6.3 with a 2.6.32-279.19.1.el6.x86_64 kernel and when it was matched
to the OS/kernel of the old hardware which was cent 5.8 with
a 2.6.18-308.11.1.el5 kernel.

Oh, now that's interesting. We've been seeing the same issue (IO stalls
on COMMIT) ond had attributed it to some bugs in the 3.2 and 3.4
kernels, partly because we don't have a credible "old server" to test
against. Now you have me wondering if there's not a hardware or driver
issue with a major HW manufacturer which just happens to be hitting
around now.

Can you detail your hardware stack so that we can compare notes?
--
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com

[Jean-David Beyer]

Post by Guillaume Smet
Steven,

Post by Steven Crandell
We saw the same performance problems when this new hardware was running
cent 6.3 with a 2.6.32-279.19.1.el6.x86_64 kernel and when it was matched
to the OS/kernel of the old hardware which was cent 5.8 with
a 2.6.18-308.11.1.el5 kernel.

Oh, now that's interesting. We've been seeing the same issue (IO stalls
on COMMIT) ond had attributed it to some bugs in the 3.2 and 3.4
kernels, partly because we don't have a credible "old server" to test
against. Now you have me wondering if there's not a hardware or driver
issue with a major HW manufacturer which just happens to be hitting
around now.
Can you detail your hardware stack so that we can compare notes?

The current Red Hat Enterprise Linux 6.4 kernel is
kernel-2.6.32-358.0.1.el6.x86_64

in case that matters.

[Steven Crandell]

Here's our hardware break down.

The logvg on the new hardware is 30MB/s slower (170 MB/s vs 200 MB/s )
than the logvg on the older hardware which was an immediately interesting
difference but we have yet to be able to create a test scenario that
successfully implicates this slower log speed in our problems. That is
something we are actively working on.


Old server hardware:
Manufacturer: Dell Inc.
Product Name: PowerEdge R810
4x Intel(R) Xeon(R) CPU E7540 @ 2.00GHz
32x16384 MB 1066 MHz DDR3
Controller 0: PERC H700 - 2 disk RAID-1 278.88 GB rootvg
Controller 1: PERC H800 - 18 disk RAID-6 2,178.00 GB datavg, 4
drive RAID-10 272.25 GB logvg, 2 hot spare
2x 278.88 GB 15K SAS on controller 0
24x 136.13 GB 15K SAS on controller 1

New server hardware:
Manufacturer: Dell Inc.
Product Name: PowerEdge R820
4x Intel(R) Xeon(R) CPU E5-4620 0 @ 2.20GHz
32x32 GB 1333 MHz DDR3
Controller 0: PERC H710P - 4 disk RAID-6 557.75 GB rootvg
Controller 1: PERC H810 - 20 disk RAID-60 4,462.00 GB datavg, 2
disk RAID-1 278.88 GB logvg, 2 hot spare
28x278.88 GB 15K SAS drives total.

Post by Steven Crandell

Post by Guillaume Smet
Steven,

Post by Steven Crandell
We saw the same performance problems when this new hardware was running
cent 6.3 with a 2.6.32-279.19.1.el6.x86_64 kernel and when it was

matched

Post by Guillaume Smet

Post by Steven Crandell
to the OS/kernel of the old hardware which was cent 5.8 with
a 2.6.18-308.11.1.el5 kernel.

Oh, now that's interesting. We've been seeing the same issue (IO stalls
on COMMIT) ond had attributed it to some bugs in the 3.2 and 3.4
kernels, partly because we don't have a credible "old server" to test
against. Now you have me wondering if there's not a hardware or driver
issue with a major HW manufacturer which just happens to be hitting
around now.
Can you detail your hardware stack so that we can compare notes?

The current Red Hat Enterprise Linux 6.4 kernel is
kernel-2.6.32-358.0.1.el6.x86_64
in case that matters.
--
http://www.postgresql.org/mailpref/pgsql-performance

[Mark Kirkwood]

Post by Steven Crandell
Here's our hardware break down.
The logvg on the new hardware is 30MB/s slower (170 MB/s vs 200 MB/s )
than the logvg on the older hardware which was an immediately interesting
difference but we have yet to be able to create a test scenario that
successfully implicates this slower log speed in our problems. That is
something we are actively working on.
Manufacturer: Dell Inc.
Product Name: PowerEdge R810
32x16384 MB 1066 MHz DDR3
Controller 0: PERC H700 - 2 disk RAID-1 278.88 GB rootvg
Controller 1: PERC H800 - 18 disk RAID-6 2,178.00 GB datavg, 4
drive RAID-10 272.25 GB logvg, 2 hot spare
2x 278.88 GB 15K SAS on controller 0
24x 136.13 GB 15K SAS on controller 1
Manufacturer: Dell Inc.
Product Name: PowerEdge R820
32x32 GB 1333 MHz DDR3
Controller 0: PERC H710P - 4 disk RAID-6 557.75 GB rootvg
Controller 1: PERC H810 - 20 disk RAID-60 4,462.00 GB datavg, 2
disk RAID-1 278.88 GB logvg, 2 hot spare
28x278.88 GB 15K SAS drives total.

Right - It is probably worth running 'pg_test_fsync' on the two logvg's
and comparing the results. This will tell you if the commit latency is
similar or not on the two disk systems.

One other difference that springs immediately to mind is that datavg is
an 18 disk RAID 6 on the old system and a 20 disk RAID 60 on the new
one...so you have about 1/2 the io performance right there.

Cheers

Mark

[John Rouillard]

Post by Steven Crandell
Here's our hardware break down.
The logvg on the new hardware is 30MB/s slower (170 MB/s vs 200 MB/s )
than the logvg on the older hardware which was an immediately interesting
difference but we have yet to be able to create a test scenario that
successfully implicates this slower log speed in our problems. That is
something we are actively working on.
Manufacturer: Dell Inc.
Product Name: PowerEdge R810
32x16384 MB 1066 MHz DDR3
Controller 0: PERC H700 - 2 disk RAID-1 278.88 GB rootvg
Controller 1: PERC H800 - 18 disk RAID-6 2,178.00 GB datavg, 4
drive RAID-10 272.25 GB logvg, 2 hot spare
2x 278.88 GB 15K SAS on controller 0
24x 136.13 GB 15K SAS on controller 1
Manufacturer: Dell Inc.
Product Name: PowerEdge R820
32x32 GB 1333 MHz DDR3
Controller 0: PERC H710P - 4 disk RAID-6 557.75 GB rootvg
Controller 1: PERC H810 - 20 disk RAID-60 4,462.00 GB datavg, 2
disk RAID-1 278.88 GB logvg, 2 hot spare
28x278.88 GB 15K SAS drives total.

Hmm, you went from a striped (raid 1/0) log volume on the old hardware
to a non-striped (raid 1) volume on the new hardware. That could
explain the speed drop. Are the disks the same speed for the two
systems?
--
-- rouilj

John Rouillard System Administrator
Renesys Corporation 603-244-9084 (cell) 603-643-9300 x 111

[Scott Marlowe]

Post by John Rouillard

Post by Steven Crandell
Here's our hardware break down.
The logvg on the new hardware is 30MB/s slower (170 MB/s vs 200 MB/s )
than the logvg on the older hardware which was an immediately interesting
difference but we have yet to be able to create a test scenario that
successfully implicates this slower log speed in our problems. That is
something we are actively working on.
Manufacturer: Dell Inc.
Product Name: PowerEdge R810
32x16384 MB 1066 MHz DDR3
Controller 0: PERC H700 - 2 disk RAID-1 278.88 GB rootvg
Controller 1: PERC H800 - 18 disk RAID-6 2,178.00 GB datavg, 4
drive RAID-10 272.25 GB logvg, 2 hot spare
2x 278.88 GB 15K SAS on controller 0
24x 136.13 GB 15K SAS on controller 1
Manufacturer: Dell Inc.
Product Name: PowerEdge R820
32x32 GB 1333 MHz DDR3
Controller 0: PERC H710P - 4 disk RAID-6 557.75 GB rootvg
Controller 1: PERC H810 - 20 disk RAID-60 4,462.00 GB datavg, 2
disk RAID-1 278.88 GB logvg, 2 hot spare
28x278.88 GB 15K SAS drives total.

Hmm, you went from a striped (raid 1/0) log volume on the old hardware
to a non-striped (raid 1) volume on the new hardware. That could
explain the speed drop. Are the disks the same speed for the two
systems?

Yeah that's a terrible tradeoff there. Just throw 4 disks in a
RAID-10 instead of RAID-60. With 4 disks you'll get the same storage
and much better performance from RAID-10.

Also consider using larger drives and a RAID-10 for your big drive
array. RAID-6 or RAID-60 is notoriously slow for databases,
especially for random access.

[Steven Crandell]

Mark,
I ran pg_fsync_test on log and data LV's on both old and new hardware.

New hardware out performed old on every measurable on the log LV

Same for the data LV's except for the 16kB open_sync write where the old
hardware edged out the new by a hair (18649 vs 17999 ops/sec)
and write, fsync, close where they were effectively tied.

Post by John Rouillard

Post by Steven Crandell
Here's our hardware break down.
The logvg on the new hardware is 30MB/s slower (170 MB/s vs 200 MB/s )
than the logvg on the older hardware which was an immediately interesting
difference but we have yet to be able to create a test scenario that
successfully implicates this slower log speed in our problems. That is
something we are actively working on.
Manufacturer: Dell Inc.
Product Name: PowerEdge R810
32x16384 MB 1066 MHz DDR3
Controller 0: PERC H700 - 2 disk RAID-1 278.88 GB rootvg
Controller 1: PERC H800 - 18 disk RAID-6 2,178.00 GB datavg, 4
drive RAID-10 272.25 GB logvg, 2 hot spare
2x 278.88 GB 15K SAS on controller 0
24x 136.13 GB 15K SAS on controller 1
Manufacturer: Dell Inc.
Product Name: PowerEdge R820
32x32 GB 1333 MHz DDR3
Controller 0: PERC H710P - 4 disk RAID-6 557.75 GB rootvg
Controller 1: PERC H810 - 20 disk RAID-60 4,462.00 GB datavg,

2

Post by Steven Crandell
disk RAID-1 278.88 GB logvg, 2 hot spare
28x278.88 GB 15K SAS drives total.

Hmm, you went from a striped (raid 1/0) log volume on the old hardware
to a non-striped (raid 1) volume on the new hardware. That could
explain the speed drop. Are the disks the same speed for the two
systems?
--
-- rouilj
John Rouillard System Administrator
Renesys Corporation 603-244-9084 (cell) 603-643-9300 x 111
--
http://www.postgresql.org/mailpref/pgsql-performance

[Scott Marlowe]

I'd be more interested in the random results from bonnie++ but my real
world experience tells me that for heavily parallel writes etc a
RAID-10 will stomp a RAID-6 or RAID-60 on the same number of drives.

On Mon, Mar 4, 2013 at 4:47 PM, Steven Crandell

Post by Steven Crandell
Mark,
I ran pg_fsync_test on log and data LV's on both old and new hardware.
New hardware out performed old on every measurable on the log LV
Same for the data LV's except for the 16kB open_sync write where the old
hardware edged out the new by a hair (18649 vs 17999 ops/sec)
and write, fsync, close where they were effectively tied.

Post by John Rouillard

Post by Steven Crandell
Here's our hardware break down.
The logvg on the new hardware is 30MB/s slower (170 MB/s vs 200 MB/s )
than the logvg on the older hardware which was an immediately interesting
difference but we have yet to be able to create a test scenario that
successfully implicates this slower log speed in our problems. That is
something we are actively working on.
Manufacturer: Dell Inc.
Product Name: PowerEdge R810
32x16384 MB 1066 MHz DDR3
Controller 0: PERC H700 - 2 disk RAID-1 278.88 GB rootvg
Controller 1: PERC H800 - 18 disk RAID-6 2,178.00 GB datavg, 4
drive RAID-10 272.25 GB logvg, 2 hot spare
2x 278.88 GB 15K SAS on controller 0
24x 136.13 GB 15K SAS on controller 1
Manufacturer: Dell Inc.
Product Name: PowerEdge R820
32x32 GB 1333 MHz DDR3
Controller 0: PERC H710P - 4 disk RAID-6 557.75 GB rootvg
Controller 1: PERC H810 - 20 disk RAID-60 4,462.00 GB datavg, 2
disk RAID-1 278.88 GB logvg, 2 hot spare
28x278.88 GB 15K SAS drives total.

Hmm, you went from a striped (raid 1/0) log volume on the old hardware
to a non-striped (raid 1) volume on the new hardware. That could
explain the speed drop. Are the disks the same speed for the two
systems?
--
-- rouilj
John Rouillard System Administrator
Renesys Corporation 603-244-9084 (cell) 603-643-9300 x 111
--
http://www.postgresql.org/mailpref/pgsql-performance

[Steven Crandell]

Scott,

Long story short, yes I agree, the raids are all kinds of wrong and if I
had been involved in the build processes they would look very different
right now.

I have been playing around with different bs= and count= settings for some
simple dd tests tonight and found some striking differences that finally
show the new hardware under performing (significantly!) when compared to
the old hardware.

That said, we are still struggling to find a postgres-specific test that
yields sufficiently different results on old and new hardware that it could
serve as an indicator that we had solved the problem prior to shoving this
box back into production service and crossing our fingers. The moment we
fix the raids, we give away a prime testing scenario. Tomorrow I plan to
split the difference and fix the raids on one of the new boxes and not the
other.

We are also working on capturing prod logs for playback but that is proving
non-trivial due to our existing performance bottleneck and
some eccentricities associated with our application.

more to come on this hopefully


many thanks for all of the insights thus far.

Post by Scott Marlowe
I'd be more interested in the random results from bonnie++ but my real
world experience tells me that for heavily parallel writes etc a
RAID-10 will stomp a RAID-6 or RAID-60 on the same number of drives.
On Mon, Mar 4, 2013 at 4:47 PM, Steven Crandell

Post by Steven Crandell
Mark,
I ran pg_fsync_test on log and data LV's on both old and new hardware.
New hardware out performed old on every measurable on the log LV
Same for the data LV's except for the 16kB open_sync write where the old
hardware edged out the new by a hair (18649 vs 17999 ops/sec)
and write, fsync, close where they were effectively tied.

Post by John Rouillard

Post by Steven Crandell
Here's our hardware break down.
The logvg on the new hardware is 30MB/s slower (170 MB/s vs 200 MB/s

)

Post by Steven Crandell

Post by John Rouillard

Post by Steven Crandell
than the logvg on the older hardware which was an immediately interesting
difference but we have yet to be able to create a test scenario that
successfully implicates this slower log speed in our problems. That is
something we are actively working on.
Manufacturer: Dell Inc.
Product Name: PowerEdge R810
32x16384 MB 1066 MHz DDR3
Controller 0: PERC H700 - 2 disk RAID-1 278.88 GB rootvg
Controller 1: PERC H800 - 18 disk RAID-6 2,178.00 GB datavg, 4
drive RAID-10 272.25 GB logvg, 2 hot spare
2x 278.88 GB 15K SAS on controller 0
24x 136.13 GB 15K SAS on controller 1
Manufacturer: Dell Inc.
Product Name: PowerEdge R820
32x32 GB 1333 MHz DDR3
Controller 0: PERC H710P - 4 disk RAID-6 557.75 GB rootvg
Controller 1: PERC H810 - 20 disk RAID-60 4,462.00 GB

datavg,

Post by Steven Crandell

Post by John Rouillard

Post by Steven Crandell
2
disk RAID-1 278.88 GB logvg, 2 hot spare
28x278.88 GB 15K SAS drives total.

Hmm, you went from a striped (raid 1/0) log volume on the old hardware
to a non-striped (raid 1) volume on the new hardware. That could
explain the speed drop. Are the disks the same speed for the two
systems?
--
-- rouilj
John Rouillard System Administrator
Renesys Corporation 603-244-9084 (cell) 603-643-9300 x 111
--
Sent via pgsql-performance mailing list (
http://www.postgresql.org/mailpref/pgsql-performance

--
To understand recursion, one must first understand recursion.

[Jesper Krogh]

Post by Steven Crandell
Recently I moved my ~600G / ~15K TPS database from a
to a newer server with
The move was from v9.1.4 to v9.1.8 (eventually also tested with v9.1.4 on the new hardware) and was done via base backup followed by slave promotion.
All postgres configurations were matched exactly as were system and kernel parameters.

my guess is that you have gone down in clockfrequency on memory when you doubled the amount of memory

in a mainly memory cached database the performance is extremely sensitive to memory speed

Jesper

[Sergey Konoplev]

On Fri, Mar 1, 2013 at 1:52 AM, Steven Crandell

Post by Steven Crandell
As far as we were able to gather in the frantic moments of downtime,
hundreds of queries were hanging up while trying to COMMIT. This in turn
caused new queries backup as they waited for locks and so on.
Given that we're dealing with new hardware and the fact that this still acts
a lot like a NUMA issue, are there other settings we should be adjusting to
deal with possible performance problems associated with NUMA?
Does this sound like something else entirely?

It does. I collected a number of kernel (and not only) tuning issues
with short explanations to prevent it from affecting database behavior
badly. Try to follow them:

https://code.google.com/p/pgcookbook/wiki/Database_Server_Configuration

--
Sergey Konoplev
Database and Software Architect
http://www.linkedin.com/in/grayhemp

Phones:
USA [phone]
Russia, Moscow [phone]
Russia, Krasnodar [phone]

Skype: gray-hemp
Jabber: ***@gmail.com