[Geoserver-users] Same Configuration, Different computer, very slow

Hi ForBode,
I would like to ask you for some details in your posting:

When you descripe your computer configuration, I see some differences:

Computer 1 has Win Server 2008 R2 64 Bit, but Computer 2 no 64 Bit?
How much memory has computer 1?
Which Tomcat version is running on Computer 2?

As already described in earlier postings bad performance often depends on
JVM memory-settings. Try to set -Xmx parameter when starting your tomcat as
described in GeoServer-documentation.

Can you tell us something about system activities/usage especially the size
of the used memory and swap file?

I think, there are two main directions where I would search:

1. Memory-settings of the JVM
2. Swapping activities/memory usage

Cheers
Stefan

--
View this message in context: http://osgeo-org.1560.n6.nabble.com/Same-Configuration-Different-computer-very-slow-tp5028863p5029767.html
Sent from the GeoServer - User mailing list archive at Nabble.com.

Sure, I appreaciate any help or feedback.

I’ll be more specific. I switched to all 32 bit so I could make use of JAI and ImageIO. Computers are as follows:

Computer 1
Hardware: Macbook Pro (latest)
HDD: SSD
OS : Windows Server 2008 R2 64 bit Enterprise

VM : Yes, Parallels
Memory : 3 gb
Tomcat : 32 bit 7.0.34
JVM : 32 bit 1.6.0_38-b05
JVM Mem Set : -Xms256m -Xmx1048m -XX:+UseParallelOldGC -XX:NewRatio=2 -XX:MaxPermSize=256m -XX:+AggressiveOpts
Tomcat Mem Set: --JvmMs=256 --JvmMx=1048
JAI : Enabled
ImageIO: Enabled
GeoCache: Enabled for gridset 26915 (Same EPSG as my layers)

Status: Runs good. Returns images in an acceptable amount of time. Does not crash. If I stress the map tiling it takes longer, but is still fast enough.

Computer 2
Hardware: Linux Server
HDD: 1.8tb raid 5 disk
OS : Windows Server 2008 R2 64 bit Enterprise

VM : Yes, ESXI Vmware
Processor: Intel Xeon 2.67 ghz with 3 processors allocated to VM

Memory : 6 gb allocated to VM
Tomcat : 32 bit 7.0.34
JVM : 32 bit 1.6.0_38-b05
JVM Mem Set : -Xms256m -Xmx1048m -XX:+UseParallelOldGC -XX:NewRatio=2 -XX:MaxPermSize=256m -XX:+AggressiveOpts
Tomcat Mem Set: --JvmMs=256 --JvmMx=1048

JAI : Enabled
ImageIO: Enabled
GeoCache: Enabled for gridset 26915 (Same EPSG as my layers)

Status: Runs good. Returns images in an acceptable amount of time. Does not crash when I stress the maps. It just takes a lot longer to deal with tiles.

Computer 3
Hardware: Dell Server
Processor: Intel Xeon 2.4 ghz with 2 processors

HDD: 7200 RPM single SATA disk
OS : Windows Server 2008 R2 64 bit Standard

VM : None
Memory : 16 gb
Tomcat : 32 bit 7.0.34
JVM : 32 bit 1.6.0_38-b05
JVM Mem Set : -Xms256m -Xmx1048m -XX:+UseParallelOldGC -XX:NewRatio=2 -XX:MaxPermSize=256m -XX:+AggressiveOpts
Tomcat Mem Set: --JvmMs=256 --JvmMx=1048

JAI : Enabled
ImageIO: Enabled
GeoCache: Enabled for gridset 26915 (Same EPSG as my layers)

Status: Runs good IF I don’t zoom to fast or pan the map to fast. The moment I do Tomcat crashes and the service stops. I always see a spike in disk usage, but larger than the other computers.

Notes on All Three Computers
All three see a major CPU drain when loading maps. A single user will spike the CPU to 90%. Each computer handles that differently, with Computer 2 being the best (recovers fast).

I am still learning GeoWebCache and believe I have a thing or two wrong. I need to make sure all my resolutions match the gridsets.

If you like I can send get you videos of the performance monitoring. I may even be able to allow you to log into Computer 2. Computer 3, however, is at a restricted site.

Other odd behavior
I have two GeoTiffs.
GeoTiff 1 - 600 mb. EPSG:26915. Loads fine. Cache’s fine. I can see the whole map in one screen very fast. Zooming works great. Disk IO is high, but CPU and network is lower. Since this layer is cached it makes sense that disk IO would be higher.

GeoTiff 2 - 500 mb. EPSG: 26915. Loads fine. Does NOT cache for some reason. When zooming in close it tiles ok, but trying to see the whole map spikes the CPU and network for long periods, but not the disk (compared to CPU or network). Since this layer isn’t cached it makes sense that there is not as much disk io (From what Andrea said).

I think that will help. Please, any advice is appreciated.


Brad A. Bode
Principal
Software Systems
Foundry Engineering


From: Stefan_E [via OSGeo.org] <[hidden email]>
To: ForBode <[hidden email]>
Sent: Thursday, January 24, 2013 11:06 PM
Subject: Re: Same Configuration, Different computer, very slow

Hi ForBode,
I would like to ask you for some details in your posting:

When you descripe your computer configuration, I see some differences:

Computer 1 has Win Server 2008 R2 64 Bit, but Computer 2 no 64 Bit?
How much memory has computer 1?
Which Tomcat version is running on Computer 2?

As already described in earlier postings bad performance often depends on JVM memory-settings. Try to set -Xmx parameter when starting your tomcat as described in GeoServer-documentation.

Can you tell us something about system activities/usage especially the size of the used memory and swap file?

I think, there are two main directions where I would search:

  1. Memory-settings of the JVM
  2. Swapping activities/memory usage

Cheers
Stefan


If you reply to this email, your message will be added to the discussion below:
http://osgeo-org.1560.n6.nabble.com/Same-Configuration-Different-computer-very-slow-tp5028863p5029767.html
To unsubscribe from Same Configuration, Different computer, very slow, click here.
NAML


View this message in context: Re: Same Configuration, Different computer, very slow
Sent from the GeoServer - User mailing list archive at Nabble.com.

On Fri, Jan 25, 2013 at 8:16 PM, ForBode <cabaal@anonymised.com> wrote:

GeoTiff 2 - 500 mb. EPSG: 26915. Loads fine. Does NOT cache for some reason. When zooming in close it tiles ok, but trying to see the whole map spikes the CPU and network for long periods, but not the disk (compared to CPU or network). Since this layer isn’t cached it makes sense that there is not as much disk io (From what Andrea said).

Hum… are you sure this layer has a proper set of overviews?
See this presentation for more info on the topic:

http://demo.geo-solutions.it/share/foss4g2011/gs_steroids_sgiannec_foss4g2011.pdf

Cheers
Andrea

==
Our support, Your Success! Visit http://opensdi.geo-solutions.it for more information.

Ing. Andrea Aime
@geowolf
Technical Lead

GeoSolutions S.A.S.
Via Poggio alle Viti 1187
55054 Massarosa (LU)
Italy
phone: +39 0584 962313
fax: +39 0584 1660272
mob: +39 339 8844549

http://www.geo-solutions.it
http://twitter.com/geosolutions_it


Hm, I am not the biggest expert, but I can't recognize big problems. Perhaps the disk performance is a bit different, but thats all.

Have you played a bit with the memory settings?

And what databases are you using?

Perhaps it would be interesting to look inside the tomcat especially for garbage collection. Do you know the jkmanager/jmx/jvm/jconsole-things?

Cheers
Stefan

-----Ursprüngliche Nachricht-----
Von: ForBode [mailto:cabaal@anonymised.com]
Gesendet: Fr 25.01.2013 20:16
An: geoserver-users@lists.sourceforge.net
Betreff: [bulk]: Re: [Geoserver-users] Same Configuration, Different computer,very slow

Sure, I appreaciate any help or feedback.

I'll be more specific. I switched to all 32 bit so I could make use of JAI and ImageIO. Computers are as follows:

Computer 1
Hardware: Macbook Pro (latest)
HDD: SSD
OS : Windows Server 2008 R2 64 bit Enterprise

VM : Yes, Parallels
Memory : 3 gb
Tomcat : 32 bit 7.0.34
JVM : 32 bit 1.6.0_38-b05
JVM Mem Set : -Xms256m -Xmx1048m -XX:+UseParallelOldGC -XX:NewRatio=2 -XX:MaxPermSize=256m -XX:+AggressiveOpts
Tomcat Mem Set: --JvmMs=256 --JvmMx=1048
JAI : Enabled
ImageIO: Enabled
GeoCache: Enabled for gridset 26915 (Same EPSG as my layers)

Status: Runs good. Returns images in an acceptable amount of time. Does not crash. If I stress the map tiling it takes longer, but is still fast enough.

Computer 2
Hardware: Linux Server
HDD: 1.8tb raid 5 disk
OS : Windows Server 2008 R2 64 bit Enterprise

VM : Yes, ESXI Vmware
Processor: Intel Xeon 2.67 ghz with 3 processors allocated to VM

Memory : 6 gb allocated to VM
Tomcat : 32 bit 7.0.34
JVM : 32 bit 1.6.0_38-b05
JVM Mem Set : -Xms256m -Xmx1048m -XX:+UseParallelOldGC -XX:NewRatio=2 -XX:MaxPermSize=256m -XX:+AggressiveOpts
Tomcat Mem Set: --JvmMs=256 --JvmMx=1048
JAI : Enabled
ImageIO: Enabled
GeoCache: Enabled for gridset 26915 (Same EPSG as my layers)

Status: Runs good. Returns images in an acceptable amount of time. Does not crash when I stress the maps. It just takes a lot longer to deal with tiles.

Computer 3
Hardware: Dell Server
Processor: Intel Xeon 2.4 ghz with 2 processors

HDD: 7200 RPM single SATA disk
OS : Windows Server 2008 R2 64 bit Standard

VM : None
Memory : 16 gb
Tomcat : 32 bit 7.0.34
JVM : 32 bit 1.6.0_38-b05
JVM Mem Set : -Xms256m -Xmx1048m -XX:+UseParallelOldGC -XX:NewRatio=2 -XX:MaxPermSize=256m -XX:+AggressiveOpts
Tomcat Mem Set: --JvmMs=256 --JvmMx=1048
JAI : Enabled
ImageIO: Enabled
GeoCache: Enabled for gridset 26915 (Same EPSG as my layers)

Status: Runs good IF I don't zoom to fast or pan the map to fast. The moment I do Tomcat crashes and the service stops. I always see a spike in disk usage, but larger than the other computers.

Notes on All Three Computers
All three see a major CPU drain when loading maps. A single user will spike the CPU to 90%. Each computer handles that differently, with Computer 2 being the best (recovers fast).

I am still learning GeoWebCache and believe I have a thing or two wrong. I need to make sure all my resolutions match the gridsets.

If you like I can send get you videos of the performance monitoring. I may even be able to allow you to log into Computer 2. Computer 3, however, is at a restricted site.

Other odd behavior
I have two GeoTiffs.
GeoTiff 1 - 600 mb. EPSG:26915. Loads fine. Cache's fine. I can see the whole map in one screen very fast. Zooming works great. Disk IO is high, but CPU and network is lower. Since this layer is cached it makes sense that disk IO would be higher.

GeoTiff 2 - 500 mb. EPSG: 26915. Loads fine. Does NOT cache for some reason. When zooming in close it tiles ok, but trying to see the whole map spikes the CPU and network for long periods, but not the disk (compared to CPU or network). Since this layer isn't cached it makes sense that there is not as much disk io (From what Andrea said).

I think that will help. Please, any advice is appreciated.

________________________________
Brad A. Bode
Principal
Software Systems
Foundry Engineering

________________________________
From: Stefan_E [via OSGeo.org] <ml-node+s1560n5029767h90@anonymised.com>
To: ForBode <cabaal@anonymised.com>
Sent: Thursday, January 24, 2013 11:06 PM
Subject: Re: Same Configuration, Different computer, very slow

Hi ForBode,
I would like to ask you for some details in your posting:

When you descripe your computer configuration, I see some differences:

Computer 1 has Win Server 2008 R2 64 Bit, but Computer 2 no 64 Bit?
How much memory has computer 1?
Which Tomcat version is running on Computer 2?

As already described in earlier postings bad performance often depends on JVM memory-settings. Try to set -Xmx parameter when starting your tomcat as described in GeoServer-documentation.

Can you tell us something about system activities/usage especially the size of the used memory and swap file?

I think, there are two main directions where I would search:

1. Memory-settings of the JVM
2. Swapping activities/memory usage

Cheers
Stefan

________________________________

If you reply to this email, your message will be added to the discussion below:
http://osgeo-org.1560.n6.nabble.com/Same-Configuration-Different-computer-very-slow-tp5028863p5029767.html
To unsubscribe from Same Configuration, Different computer, very slow, click here.
NAML

--
View this message in context: http://osgeo-org.1560.n6.nabble.com/Same-Configuration-Different-computer-very-slow-tp5028863p5029968.html
Sent from the GeoServer - User mailing list archive at Nabble.com.

Andrea, thank you so very much.
It has been hectic trying to sort through what to turn on and off for raster data. In addition to that my background is not in image processing at all. The article you posted was enlightening. The results:

The GeoTiff that rendered faster did indeed have overviews.

The GeoTiff that locked up the CPU did NOT.

I turned off all caching (I’d rather not use it) and loaded the corrected GeoTiffs (I used Gdal) and it’s blazingly fast now on my Mac Pro. I will be trying it on my Computer 2 (Server) later and hope to see even better results.

So my question is… should I even bother with caching? It doesn’t seem that it would improve performance. In fact, on my Computer 3 (production at site) disk thrashing seems to kill tomcat. I hope that with this new GeoTiff and no caching it will remedy the thrashing. After all, I’d rather use more CPU, in this case, than disk.

Thoughts?


Brad A. Bode
Principal
Software Systems
Foundry Engineering


View this message in context: Re: Same Configuration, Different computer, very slow
Sent from the GeoServer - User mailing list archive at Nabble.com.

On Sat, Jan 26, 2013 at 12:06 AM, ForBode <cabaal@anonymised.com> wrote:

Andrea, thank you so very much.
It has been hectic trying to sort through what to turn on and off for raster data. In addition to that my background is not in image processing at all. The article you posted was enlightening. The results:

The GeoTiff that rendered faster did indeed have overviews.

The GeoTiff that locked up the CPU did NOT.

I turned off all caching (I’d rather not use it) and loaded the corrected GeoTiffs (I used Gdal) and it’s blazingly fast now on my Mac Pro. I will be trying it on my Computer 2 (Server) later and hope to see even better results.

So my question is… should I even bother with caching? It doesn’t seem that it would improve performance. In fact, on my Computer 3 (production at site) disk thrashing seems to kill tomcat. I hope that with this new GeoTiff and no caching it will remedy the thrashing. After all, I’d rather use more CPU, in this case, than disk.

Hum… good question. Generally speaking caching helps more maps based on vector data than on raster data,
but still, I have seen speedup factors or 5-10 times even on raster data.
However, I’ve seen them on Linux machines running directly on the hardware (no VM), it may be that
on your server this does not apply

Cheers
Andrea

==
Our support, Your Success! Visit http://opensdi.geo-solutions.it for more information.

Ing. Andrea Aime
@geowolf
Technical Lead

GeoSolutions S.A.S.
Via Poggio alle Viti 1187
55054 Massarosa (LU)
Italy
phone: +39 0584 962313
fax: +39 0584 1660272
mob: +39 339 8844549

http://www.geo-solutions.it
http://twitter.com/geosolutions_it