[SOLVED] AW: Performance difference between servers

In short: setting the energy options on the host to „high performance“ solved the issue.

Side note: The energy option has impact only on server #2 and #3, not on #1 (see below).

In detail: BIOS RAID 1, different processors, operatings systems and a mix of fixed and dynamic virtuals disks did not explain a factor of 2 … 3 in performance. Changing the virtual disk from dynamic to fixed, different settings in postgresql.conf, changing memory allocation for the virtual guests did not show any significant improvements – the oldest server with the slowest processor, minimum RAM and slow disks was still performing better than the others.

All hosts were running on the energy option „balanced“ with no modification. However, the processor in server #1 was always running at 3.4 GHz and did not reduce clock speed in idle mode. Server #2 and #3 were running at 0.8 GHz in idle mode.

The real issue was that server #2 and #3 did not speed up when the database operation started – not to nominal speed and clearly not entering turbo mode.

Solution for server #2 and #3:

1. activate high performance mode => this causes the processor to run at nominal speed at all times

2. reduce the setting for processor minimum performance back to 5%

3. activate high performance mode again => this re-enables the dynamic clock speed but ensures that the processor speeds up when needed

Why is this not needed for server #1?

I believe that the processor running in server #1 is simply too old to reduce the clock speed. Perhaps a missing or outdated chipset driver is also somehow involved (that’s just a guess).

Thanks to everybody providing input

Klaus

Von: Rui DeSousa [mailto:rui(dot)desousa(at)icloud(dot)com]
Gesendet: Donnerstag, 16. November 2017 00:43
An: kpi6288(at)gmail(dot)com
Cc: pgsql-admin(at)postgresql(dot)org
Betreff: Re: [ADMIN] Performance difference between servers

There is a log of variables there. Is the i7 single core performance faster than the E3? Bus/memory speeds, etc. I would also consider any vCPU imbalances. You could end up with guest scheduling issue; i.e if guest requires 1 core but is allocated 4 cores; then for 4 cores still need to be available and ready to schedule the guest even if it current need is only 1 core at the moment.

On Nov 15, 2017, at 12:43 PM, kpi6288(at)gmail(dot)com <mailto:kpi6288(at)gmail(dot)com> wrote:

Trying as best as a can:

#1 HP 8200 Elite microtower, i7-2600, 16GB RAM, 1 SSD 480 GB, 1 SSD 900 GB, Hyper-V 2012 R2 (not 2008 R2 as written in my original question)

#2 Lenovo TS 150, E3-1245 V5, 40GB RAM, 2 SSD 850GB as RAID 1 (BIOS level), 2 HDDs 1TB as RAID 1 (BIOS level)

#3 model unknown, E3-1270 V3, 32GB RAM, 2 SSD 480GB as RAID 1 (BIOS level)

Server #1 is more a PC than a server.

Server #1 runs in total 4 virtual guests. 8GB and 6 „cores“ dedicated to the server running PostgreSQL and 3GB to other machines which are almost 100% idle.

Server #2 has currently on the one guets running PostgreSQL with 16GB RAM and 6 „cores“ dedicated.

Server #3 runs 2 virtual guests. 12GB and 4 cores dedicated tot he guest runnning PostgreSQL. 4 GB and 3 cores decicated to another guest with is practically idle.

Servers #2 and #3 have still plenty of free disk space while server #1 is using approx. 70% oft he disks.

Server #1 has much older SSDs than #2. Not sure about #3.

Von: Scott Whitney [ <mailto:scott(at)journyx(dot)com> mailto:scott(at)journyx(dot)com]
Gesendet: Mittwoch, 15. November 2017 18:19
An: <mailto:kpi6288(at)gmail(dot)com> kpi6288(at)gmail(dot)com
Betreff: Re: [ADMIN] Performance difference between servers

Can you post the hardware host specs for #1, #2 and #3? If those ARE the hardware specs (for example, you list SSDs), can you list the resources assigned to the guests?

_____

From: <mailto:kpi6288(at)gmail(dot)com> kpi6288(at)gmail(dot)com < <mailto:kpi6288(at)gmail(dot)com> kpi6288(at)gmail(dot)com>
Sent: Wednesday, November 15, 2017 11:15 AM
To: Scott Whitney
Subject: AW: [ADMIN] Performance difference between servers

Yes, exactly:

Three different (hardware) machines, each having a hyper-V host and the Windows Servers as Hyper-V guests.

Von: Scott Whitney [ <mailto:scott(at)journyx(dot)com> mailto:scott(at)journyx(dot)com]
Gesendet: Mittwoch, 15. November 2017 18:11
An: <mailto:kpi6288(at)gmail(dot)com> kpi6288(at)gmail(dot)com; <mailto:pgsql-admin(at)postgresql(dot)org> pgsql-admin(at)postgresql(dot)org
Betreff: Re: [ADMIN] Performance difference between servers

Are you saying that each of these servers is a virtual HyperV machine running under a hardware host?

_____

From: <mailto:pgsql-admin-owner(at)postgresql(dot)org> pgsql-admin-owner(at)postgresql(dot)org < <mailto:pgsql-admin-owner(at)postgresql(dot)org> pgsql-admin-owner(at)postgresql(dot)org> on behalf of <mailto:kpi6288(at)gmail(dot)com> kpi6288(at)gmail(dot)com < <mailto:kpi6288(at)gmail(dot)com> kpi6288(at)gmail(dot)com>
Sent: Wednesday, November 15, 2017 11:06 AM
To: <mailto:pgsql-admin(at)postgresql(dot)org> pgsql-admin(at)postgresql(dot)org
Subject: [ADMIN] Performance difference between servers

We have 3 different servers, one (the oldest) ist significantly faster at certain operations and I can’t find out the reason:

(table created in monospace font)

Server Processor RAM CrystalDiskMark pgbench CREATE DATABASE OS

#1 i7-2600 8GB 495 / 473 26.000 4.5 … 6 s W/ Server 2008 R2 on Hyper V Server 2008

#2 E3-1245 16GB 797 / 514 50.000 13 … 19 s W/ Server 2016 Std. on Hyper V Server 2016

#3 E3-1270 12GB 933 / 334 40.000 13 … 20 s W/ Server 2008 Essentials on Hyper V Server 2016

CrystalDiskMark: „Seq Q32T1“

Pgbench:

pgbench -i -s 200 -U postgres pgbench

pgbench -c 40 -j 10 -T 150 -U postgres -S -P 5 pgbench

CREATE DATABASE:

create database test with template myTemplate

All running PostgreSQL 9.6 x64 with similar configuration. All running as virtual machines, no other virtual machines creating any load. Using the same template database on all machines.

Although the indicative numbers of pgbench and CrystalDiskMark show better values for server #2 and #3, server #1 creates a new database considerably faster.

All servers have SSDs. Server #1 has two single SSDs (a system disk and a data disk). Servers #2 and #3 have only one logical disk for system and data, and this is two SSDs running as RAID 1. This may be slower but if it is, I would excpect this also to show up in the pgbench results.

The reason behind this exercise is that a test application was also reported to be slower on server #2 and 3.

Any idea how to track down the performance difference?

Thank you

Klaus

Journyx, Inc.

7600 Burnet Road #300
Austin, TX 78757
<http://www.journyx.com/> www.journyx.com

p 512.834.8888

f 512-834-8858

Do you receive our promotional emails? Click <http://journyx.com/communication-preferences> here or visit <http://journyx.com/communication-preferences> http://journyx.com/communication-preferences to unsubscribe.