Re: [Fwd: Intermine recommended java settings?]

classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Intermine recommended java settings?]

Sergio Contrino
dear colin,
i have no great experience in tomcat tuning, and hopefully justin in our
group will be able to be more useful.
i just note that usually we set GC old style with
-XX:+UseParallelGC
and that we sometime use
-XX:SoftRefLRUPolicyMSPerMB=30
for soft references.
please share any interesting result from your testing!
thanks
sergio


> ---------------------------- Original Message ----------------------------
> Subject: Intermine recommended java settings?
> From:    "Colin" <[hidden email]>
> Date:    Mon, September 28, 2015 3:52 pm
> To:      "Julie Sullivan" <[hidden email]>
> --------------------------------------------------------------------------
>
> Hi Julie
>
> I have been trying to debug the cpu spikyness on our servers
>
> I kind of feel like it is just a resource allocation issue, so I have been
> increasing our resource usage and trying to tune the servers, but it might
> also involve some java server settings
>
> Currently our production settings are this
>
> export CATALINA_OPTS="-Xms20g -Xmx20g
>    -XX:+CMSClassUnloadingEnabled
>    -XX:+CMSPermGenSweepingEnabled
>    -XX:+UseConcMarkSweepGC
>    -XX:PermSize=2g
>    -XX:MaxPermSize=2g
>    -Dorg.apache.el.parser.SKIP_IDENTIFIER_CHECK=true
>    -Dcom.sun.management.jmxremote
>    -Dcom.sun.management.jmxremote.port=8999
>    -Dcom.sun.management.jmxremote.ssl=false
>    -Dcom.sun.management.jmxremote.authenticate=false
>    -Djava.rmi.server.hostname=localhost"
>
>
> We have a 5.6 gig search index so that is pretty big, but it's still kind
> of intensive on our servers. I was considering disabling the concurrent
> mark sweep because it literally just takes a lot of cpu (like, 10+cores)
>
>
> Thanks!
>
> -Colin
>

--
sergio contrino                  InterMine, University of Cambridge
https://sergiocontrino.github.io           http://www.intermine.org

_______________________________________________
dev mailing list
[hidden email]
http://mail.intermine.org/cgi-bin/mailman/listinfo/dev
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Intermine recommended java settings?]

Colin
Thanks. I'll definitely report back on any changes that help!



-Colin

On Tue, Sep 29, 2015 at 3:44 AM, sergio contrino <[hidden email]> wrote:
dear colin,
i have no great experience in tomcat tuning, and hopefully justin in our group will be able to be more useful.
i just note that usually we set GC old style with
-XX:+UseParallelGC
and that we sometime use
-XX:SoftRefLRUPolicyMSPerMB=30
for soft references.
please share any interesting result from your testing!
thanks
sergio


---------------------------- Original Message ----------------------------
Subject: Intermine recommended java settings?
From:    "Colin" <[hidden email]>
Date:    Mon, September 28, 2015 3:52 pm
To:      "Julie Sullivan" <[hidden email]>
--------------------------------------------------------------------------

Hi Julie

I have been trying to debug the cpu spikyness on our servers

I kind of feel like it is just a resource allocation issue, so I have been
increasing our resource usage and trying to tune the servers, but it might
also involve some java server settings

Currently our production settings are this

export CATALINA_OPTS="-Xms20g -Xmx20g
   -XX:+CMSClassUnloadingEnabled
   -XX:+CMSPermGenSweepingEnabled
   -XX:+UseConcMarkSweepGC
   -XX:PermSize=2g
   -XX:MaxPermSize=2g
   -Dorg.apache.el.parser.SKIP_IDENTIFIER_CHECK=true
   -Dcom.sun.management.jmxremote
   -Dcom.sun.management.jmxremote.port=8999
   -Dcom.sun.management.jmxremote.ssl=false
   -Dcom.sun.management.jmxremote.authenticate=false
   -Djava.rmi.server.hostname=localhost"


We have a 5.6 gig search index so that is pretty big, but it's still kind
of intensive on our servers. I was considering disabling the concurrent
mark sweep because it literally just takes a lot of cpu (like, 10+cores)


Thanks!

-Colin


--
sergio contrino                  InterMine, University of Cambridge
https://sergiocontrino.github.io           http://www.intermine.org


_______________________________________________
dev mailing list
[hidden email]
http://mail.intermine.org/cgi-bin/mailman/listinfo/dev
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Intermine recommended java settings?]

Justin Clark-Casey
Hi Colin.  I haven't been working on InterMine for long and previously I hadn't worked with Java for about 7 years, though I was working with .NET VMs (mainly
Mono) during that time.  My own mine (synbiomine) is quite small so I haven't hit these issues yet, though I know we have seen some CPU spikiness with bigger
mines.  Here are some general thoughts/questions.

* Which version of Tomcat, the JVM and InterMine are you using?

* Why set -XX:+CMSClassUnloadingEnabled?  From doc, I see that it allows the JVM to unload classes that are no longer referenced.  However, I shouldn't think
this is a significant factor in InterMine/Tomcat memory usage (unless you've seen differently?) and you might save some cpu time and be on more tested JVM code
paths if you didn't have this.

* Why set -XX:PermSize=2g and -XX:MaxPermSize=2g?  I see this refers to the size of the permanent generation.  However, I would think it's better if the JVM is
left to manage this, and it shouldn't have an impact on GC since the permanent generation is not garabage collected (though maybe -XX:+CMSPermGenSweepingEnabled
has this effect, though good doc on it is hard to find).  Have you seen PermGen out of memory errors in the past?

* -XX:+UseConcMarkSweepGC does use a GC where background threads can perform collection.  In theory, the default parallel GC would see a much bigger InterMine
response pause if it needs to pause and do a lot of collection (but as ever, one should always confirm such things with measurements).  I'm surprised you see it
using so many cores though.  Is this a practically permanent thing?

* Have you measured GC activity?  It looks like the JVM has some good tools for this.  If so, is this correlated with your CPU spikes?

* Search is one thing I hope to look at soon.  InterMine is currently using a pretty old Lucene (v3) and upgrading is not straightforward since it's quite a
large amount of code and uses a third party library (Bobo facet search) which is no longer maintained.  However, one thing I want to look at is whether we can
load sections of the search index off disk as required rather than into memory.  This might help with your memory usage.

Best Regards,

--
Justin Clark-Casey, Synbiomine/InterMine Software Developer
http://synbiomine.org
http://twitter.com/justincc

On 30/09/15 00:07, Colin wrote:

> Thanks. I'll definitely report back on any changes that help!
>
>
>
> -Colin
>
> On Tue, Sep 29, 2015 at 3:44 AM, sergio contrino <[hidden email] <mailto:[hidden email]>> wrote:
>
>     dear colin,
>     i have no great experience in tomcat tuning, and hopefully justin in our group will be able to be more useful.
>     i just note that usually we set GC old style with
>     -XX:+UseParallelGC
>     and that we sometime use
>     -XX:SoftRefLRUPolicyMSPerMB=30
>     for soft references.
>     please share any interesting result from your testing!
>     thanks
>     sergio
>
>
>         ---------------------------- Original Message ----------------------------
>         Subject: Intermine recommended java settings?
>         From:    "Colin" <[hidden email] <mailto:[hidden email]>>
>         Date:    Mon, September 28, 2015 3:52 pm
>         To:      "Julie Sullivan" <[hidden email] <mailto:[hidden email]>>
>         --------------------------------------------------------------------------
>
>         Hi Julie
>
>         I have been trying to debug the cpu spikyness on our servers
>
>         I kind of feel like it is just a resource allocation issue, so I have been
>         increasing our resource usage and trying to tune the servers, but it might
>         also involve some java server settings
>
>         Currently our production settings are this
>
>         export CATALINA_OPTS="-Xms20g -Xmx20g
>             -XX:+CMSClassUnloadingEnabled
>             -XX:+CMSPermGenSweepingEnabled
>             -XX:+UseConcMarkSweepGC
>             -XX:PermSize=2g
>             -XX:MaxPermSize=2g
>             -Dorg.apache.el.parser.SKIP_IDENTIFIER_CHECK=true
>             -Dcom.sun.management.jmxremote
>             -Dcom.sun.management.jmxremote.port=8999
>             -Dcom.sun.management.jmxremote.ssl=false
>             -Dcom.sun.management.jmxremote.authenticate=false
>             -Djava.rmi.server.hostname=localhost"
>
>
>         We have a 5.6 gig search index so that is pretty big, but it's still kind
>         of intensive on our servers. I was considering disabling the concurrent
>         mark sweep because it literally just takes a lot of cpu (like, 10+cores)
>
>
>         Thanks!
>
>         -Colin
>
>
>     --
>     sergio contrino                  InterMine, University of Cambridge
>     https://sergiocontrino.github.io http://www.intermine.org
>
>

_______________________________________________
dev mailing list
[hidden email]
http://mail.intermine.org/cgi-bin/mailman/listinfo/dev
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Intermine recommended java settings?]

Colin
Hi Justin,

I think enabling GC logging is definitely the next step. We are trying to minimize redeploys but I will check into it soon.


As for other things

>* Which version of Tomcat, the JVM and InterMine are you using?

We currently have intermine 1.4.2 (looking at keeping more current) and JVM 7, Tomcat 7.0.57


>* Why set -XX:PermSize=2g and -XX:MaxPermSize=2g?

Yes we've had a lot of permgen errors in the past, generally during redeploys but we haven't had as many problems with this for awhile.


Thanks for the tips!


-Colin

On Thu, Oct 1, 2015 at 7:18 AM, Justin Clark-Casey <[hidden email]> wrote:
Hi Colin.  I haven't been working on InterMine for long and previously I hadn't worked with Java for about 7 years, though I was working with .NET VMs (mainly Mono) during that time.  My own mine (synbiomine) is quite small so I haven't hit these issues yet, though I know we have seen some CPU spikiness with bigger mines.  Here are some general thoughts/questions.

* Which version of Tomcat, the JVM and InterMine are you using?

* Why set -XX:+CMSClassUnloadingEnabled?  From doc, I see that it allows the JVM to unload classes that are no longer referenced.  However, I shouldn't think this is a significant factor in InterMine/Tomcat memory usage (unless you've seen differently?) and you might save some cpu time and be on more tested JVM code paths if you didn't have this.

* Why set -XX:PermSize=2g and -XX:MaxPermSize=2g?  I see this refers to the size of the permanent generation.  However, I would think it's better if the JVM is left to manage this, and it shouldn't have an impact on GC since the permanent generation is not garabage collected (though maybe -XX:+CMSPermGenSweepingEnabled has this effect, though good doc on it is hard to find).  Have you seen PermGen out of memory errors in the past?

* -XX:+UseConcMarkSweepGC does use a GC where background threads can perform collection.  In theory, the default parallel GC would see a much bigger InterMine response pause if it needs to pause and do a lot of collection (but as ever, one should always confirm such things with measurements).  I'm surprised you see it using so many cores though.  Is this a practically permanent thing?

* Have you measured GC activity?  It looks like the JVM has some good tools for this.  If so, is this correlated with your CPU spikes?

* Search is one thing I hope to look at soon.  InterMine is currently using a pretty old Lucene (v3) and upgrading is not straightforward since it's quite a large amount of code and uses a third party library (Bobo facet search) which is no longer maintained.  However, one thing I want to look at is whether we can load sections of the search index off disk as required rather than into memory.  This might help with your memory usage.

Best Regards,

--
Justin Clark-Casey, Synbiomine/InterMine Software Developer
http://synbiomine.org
http://twitter.com/justincc

On 30/09/15 00:07, Colin wrote:
Thanks. I'll definitely report back on any changes that help!



-Colin

On Tue, Sep 29, 2015 at 3:44 AM, sergio contrino <[hidden email] <mailto:[hidden email]>> wrote:

    dear colin,
    i have no great experience in tomcat tuning, and hopefully justin in our group will be able to be more useful.
    i just note that usually we set GC old style with
    -XX:+UseParallelGC
    and that we sometime use
    -XX:SoftRefLRUPolicyMSPerMB=30
    for soft references.
    please share any interesting result from your testing!
    thanks
    sergio


        ---------------------------- Original Message ----------------------------
        Subject: Intermine recommended java settings?
        From:    "Colin" <[hidden email] <mailto:[hidden email]>>
        Date:    Mon, September 28, 2015 3:52 pm
        To:      "Julie Sullivan" <[hidden email] <mailto:[hidden email]>>
        --------------------------------------------------------------------------

        Hi Julie

        I have been trying to debug the cpu spikyness on our servers

        I kind of feel like it is just a resource allocation issue, so I have been
        increasing our resource usage and trying to tune the servers, but it might
        also involve some java server settings

        Currently our production settings are this

        export CATALINA_OPTS="-Xms20g -Xmx20g
            -XX:+CMSClassUnloadingEnabled
            -XX:+CMSPermGenSweepingEnabled
            -XX:+UseConcMarkSweepGC
            -XX:PermSize=2g
            -XX:MaxPermSize=2g
            -Dorg.apache.el.parser.SKIP_IDENTIFIER_CHECK=true
            -Dcom.sun.management.jmxremote
            -Dcom.sun.management.jmxremote.port=8999
            -Dcom.sun.management.jmxremote.ssl=false
            -Dcom.sun.management.jmxremote.authenticate=false
            -Djava.rmi.server.hostname=localhost"


        We have a 5.6 gig search index so that is pretty big, but it's still kind
        of intensive on our servers. I was considering disabling the concurrent
        mark sweep because it literally just takes a lot of cpu (like, 10+cores)


        Thanks!

        -Colin


    --
    sergio contrino                  InterMine, University of Cambridge
    https://sergiocontrino.github.io http://www.intermine.org




_______________________________________________
dev mailing list
[hidden email]
http://mail.intermine.org/cgi-bin/mailman/listinfo/dev
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Intermine recommended java settings?]

Colin
Just thought I'd update. It looks like just increasing the memory for tomcat a little more removed issues.

>However, one thing I want to look at is whether we can load sections of the search index off disk as required rather than into memory

Looks like this was the key here. I guess this was sort of a misunderstanding for me that the search index was fully a tomcat thing and not a database thing, I sort of thought that if our databases were tuned properly then the tomcat would function well, but adding more resources to tomcat so that it could handle the search index fully in memory seems to be the key to fixing this problem!

Now there's no spiking and the webapp is a lot faster now too.



-Colin


On Fri, Oct 2, 2015 at 11:28 AM, Colin <[hidden email]> wrote:
Hi Justin,

I think enabling GC logging is definitely the next step. We are trying to minimize redeploys but I will check into it soon.


As for other things

>* Which version of Tomcat, the JVM and InterMine are you using?

We currently have intermine 1.4.2 (looking at keeping more current) and JVM 7, Tomcat 7.0.57


>* Why set -XX:PermSize=2g and -XX:MaxPermSize=2g?

Yes we've had a lot of permgen errors in the past, generally during redeploys but we haven't had as many problems with this for awhile.


Thanks for the tips!


-Colin

On Thu, Oct 1, 2015 at 7:18 AM, Justin Clark-Casey <[hidden email]> wrote:
Hi Colin.  I haven't been working on InterMine for long and previously I hadn't worked with Java for about 7 years, though I was working with .NET VMs (mainly Mono) during that time.  My own mine (synbiomine) is quite small so I haven't hit these issues yet, though I know we have seen some CPU spikiness with bigger mines.  Here are some general thoughts/questions.

* Which version of Tomcat, the JVM and InterMine are you using?

* Why set -XX:+CMSClassUnloadingEnabled?  From doc, I see that it allows the JVM to unload classes that are no longer referenced.  However, I shouldn't think this is a significant factor in InterMine/Tomcat memory usage (unless you've seen differently?) and you might save some cpu time and be on more tested JVM code paths if you didn't have this.

* Why set -XX:PermSize=2g and -XX:MaxPermSize=2g?  I see this refers to the size of the permanent generation.  However, I would think it's better if the JVM is left to manage this, and it shouldn't have an impact on GC since the permanent generation is not garabage collected (though maybe -XX:+CMSPermGenSweepingEnabled has this effect, though good doc on it is hard to find).  Have you seen PermGen out of memory errors in the past?

* -XX:+UseConcMarkSweepGC does use a GC where background threads can perform collection.  In theory, the default parallel GC would see a much bigger InterMine response pause if it needs to pause and do a lot of collection (but as ever, one should always confirm such things with measurements).  I'm surprised you see it using so many cores though.  Is this a practically permanent thing?

* Have you measured GC activity?  It looks like the JVM has some good tools for this.  If so, is this correlated with your CPU spikes?

* Search is one thing I hope to look at soon.  InterMine is currently using a pretty old Lucene (v3) and upgrading is not straightforward since it's quite a large amount of code and uses a third party library (Bobo facet search) which is no longer maintained.  However, one thing I want to look at is whether we can load sections of the search index off disk as required rather than into memory.  This might help with your memory usage.

Best Regards,

--
Justin Clark-Casey, Synbiomine/InterMine Software Developer
http://synbiomine.org
http://twitter.com/justincc

On 30/09/15 00:07, Colin wrote:
Thanks. I'll definitely report back on any changes that help!



-Colin

On Tue, Sep 29, 2015 at 3:44 AM, sergio contrino <[hidden email] <mailto:[hidden email]>> wrote:

    dear colin,
    i have no great experience in tomcat tuning, and hopefully justin in our group will be able to be more useful.
    i just note that usually we set GC old style with
    -XX:+UseParallelGC
    and that we sometime use
    -XX:SoftRefLRUPolicyMSPerMB=30
    for soft references.
    please share any interesting result from your testing!
    thanks
    sergio


        ---------------------------- Original Message ----------------------------
        Subject: Intermine recommended java settings?
        From:    "Colin" <[hidden email] <mailto:[hidden email]>>
        Date:    Mon, September 28, 2015 3:52 pm
        To:      "Julie Sullivan" <[hidden email] <mailto:[hidden email]>>
        --------------------------------------------------------------------------

        Hi Julie

        I have been trying to debug the cpu spikyness on our servers

        I kind of feel like it is just a resource allocation issue, so I have been
        increasing our resource usage and trying to tune the servers, but it might
        also involve some java server settings

        Currently our production settings are this

        export CATALINA_OPTS="-Xms20g -Xmx20g
            -XX:+CMSClassUnloadingEnabled
            -XX:+CMSPermGenSweepingEnabled
            -XX:+UseConcMarkSweepGC
            -XX:PermSize=2g
            -XX:MaxPermSize=2g
            -Dorg.apache.el.parser.SKIP_IDENTIFIER_CHECK=true
            -Dcom.sun.management.jmxremote
            -Dcom.sun.management.jmxremote.port=8999
            -Dcom.sun.management.jmxremote.ssl=false
            -Dcom.sun.management.jmxremote.authenticate=false
            -Djava.rmi.server.hostname=localhost"


        We have a 5.6 gig search index so that is pretty big, but it's still kind
        of intensive on our servers. I was considering disabling the concurrent
        mark sweep because it literally just takes a lot of cpu (like, 10+cores)


        Thanks!

        -Colin


    --
    sergio contrino                  InterMine, University of Cambridge
    https://sergiocontrino.github.io http://www.intermine.org





_______________________________________________
dev mailing list
[hidden email]
http://mail.intermine.org/cgi-bin/mailman/listinfo/dev
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Intermine recommended java settings?]

Justin Clark-Casey
Great to hear that, Colin.  Yeah, the search index is stuffed into the database as a blob at the data build stage (in the intermine_metadata table).  On first
search it's rehydrated into memory and beyond that there's no further database interaction.

Interesting to hear that this was causing you an issue.  All the more reason to look at whether we can stop doing this.

--
Justin Clark-Casey, Synbiomine/InterMine Software Developer
http://synbiomine.org
http://twitter.com/justincc

On 04/10/15 18:35, Colin wrote:

> Just thought I'd update. It looks like just increasing the memory for tomcat a little more removed issues.
>
>  >However, one thing I want to look at is whether we can load sections of the search index off disk as required rather than into memory
>
> Looks like this was the key here. I guess this was sort of a misunderstanding for me that the search index was fully a tomcat thing and not a database thing, I
> sort of thought that if our databases were tuned properly then the tomcat would function well, but adding more resources to tomcat so that it could handle the
> search index fully in memory seems to be the key to fixing this problem!
>
> Now there's no spiking and the webapp is a lot faster now too.
>
>
>
> -Colin
>
>
> On Fri, Oct 2, 2015 at 11:28 AM, Colin <[hidden email] <mailto:[hidden email]>> wrote:
>
>     Hi Justin,
>
>     I think enabling GC logging is definitely the next step. We are trying to minimize redeploys but I will check into it soon.
>
>
>     As for other things
>
>     >* Which version of Tomcat, the JVM and InterMine are you using?
>
>     We currently have intermine 1.4.2 (looking at keeping more current) and JVM 7, Tomcat 7.0.57
>
>
>     >* Why set -XX:PermSize=2g and -XX:MaxPermSize=2g?
>
>     Yes we've had a lot of permgen errors in the past, generally during redeploys but we haven't had as many problems with this for awhile.
>
>
>     Thanks for the tips!
>
>
>     -Colin
>
>     On Thu, Oct 1, 2015 at 7:18 AM, Justin Clark-Casey <[hidden email] <mailto:[hidden email]>> wrote:
>
>         Hi Colin.  I haven't been working on InterMine for long and previously I hadn't worked with Java for about 7 years, though I was working with .NET VMs
>         (mainly Mono) during that time.  My own mine (synbiomine) is quite small so I haven't hit these issues yet, though I know we have seen some CPU
>         spikiness with bigger mines.  Here are some general thoughts/questions.
>
>         * Which version of Tomcat, the JVM and InterMine are you using?
>
>         * Why set -XX:+CMSClassUnloadingEnabled?  From doc, I see that it allows the JVM to unload classes that are no longer referenced.  However, I shouldn't
>         think this is a significant factor in InterMine/Tomcat memory usage (unless you've seen differently?) and you might save some cpu time and be on more
>         tested JVM code paths if you didn't have this.
>
>         * Why set -XX:PermSize=2g and -XX:MaxPermSize=2g?  I see this refers to the size of the permanent generation.  However, I would think it's better if the
>         JVM is left to manage this, and it shouldn't have an impact on GC since the permanent generation is not garabage collected (though maybe
>         -XX:+CMSPermGenSweepingEnabled has this effect, though good doc on it is hard to find).  Have you seen PermGen out of memory errors in the past?
>
>         * -XX:+UseConcMarkSweepGC does use a GC where background threads can perform collection.  In theory, the default parallel GC would see a much bigger
>         InterMine response pause if it needs to pause and do a lot of collection (but as ever, one should always confirm such things with measurements).  I'm
>         surprised you see it using so many cores though.  Is this a practically permanent thing?
>
>         * Have you measured GC activity?  It looks like the JVM has some good tools for this.  If so, is this correlated with your CPU spikes?
>
>         * Search is one thing I hope to look at soon.  InterMine is currently using a pretty old Lucene (v3) and upgrading is not straightforward since it's
>         quite a large amount of code and uses a third party library (Bobo facet search) which is no longer maintained.  However, one thing I want to look at is
>         whether we can load sections of the search index off disk as required rather than into memory.  This might help with your memory usage.
>
>         Best Regards,
>
>         --
>         Justin Clark-Casey, Synbiomine/InterMine Software Developer
>         http://synbiomine.org
>         http://twitter.com/justincc
>
>         On 30/09/15 00:07, Colin wrote:
>
>             Thanks. I'll definitely report back on any changes that help!
>
>
>
>             -Colin
>
>             On Tue, Sep 29, 2015 at 3:44 AM, sergio contrino <[hidden email] <mailto:[hidden email]> <mailto:[hidden email]
>             <mailto:[hidden email]>>> wrote:
>
>                  dear colin,
>                  i have no great experience in tomcat tuning, and hopefully justin in our group will be able to be more useful.
>                  i just note that usually we set GC old style with
>                  -XX:+UseParallelGC
>                  and that we sometime use
>                  -XX:SoftRefLRUPolicyMSPerMB=30
>                  for soft references.
>                  please share any interesting result from your testing!
>                  thanks
>                  sergio
>
>
>                      ---------------------------- Original Message ----------------------------
>                      Subject: Intermine recommended java settings?
>                      From:    "Colin" <[hidden email] <mailto:[hidden email]> <mailto:[hidden email] <mailto:[hidden email]>>>
>                      Date:    Mon, September 28, 2015 3:52 pm
>                      To:      "Julie Sullivan" <[hidden email] <mailto:[hidden email]> <mailto:[hidden email] <mailto:[hidden email]>>>
>                      --------------------------------------------------------------------------
>
>                      Hi Julie
>
>                      I have been trying to debug the cpu spikyness on our servers
>
>                      I kind of feel like it is just a resource allocation issue, so I have been
>                      increasing our resource usage and trying to tune the servers, but it might
>                      also involve some java server settings
>
>                      Currently our production settings are this
>
>                      export CATALINA_OPTS="-Xms20g -Xmx20g
>                          -XX:+CMSClassUnloadingEnabled
>                          -XX:+CMSPermGenSweepingEnabled
>                          -XX:+UseConcMarkSweepGC
>                          -XX:PermSize=2g
>                          -XX:MaxPermSize=2g
>                          -Dorg.apache.el.parser.SKIP_IDENTIFIER_CHECK=true
>                          -Dcom.sun.management.jmxremote
>                          -Dcom.sun.management.jmxremote.port=8999
>                          -Dcom.sun.management.jmxremote.ssl=false
>                          -Dcom.sun.management.jmxremote.authenticate=false
>                          -Djava.rmi.server.hostname=localhost"
>
>
>                      We have a 5.6 gig search index so that is pretty big, but it's still kind
>                      of intensive on our servers. I was considering disabling the concurrent
>                      mark sweep because it literally just takes a lot of cpu (like, 10+cores)
>
>
>                      Thanks!
>
>                      -Colin
>
>
>                  --
>                  sergio contrino                  InterMine, University of Cambridge
>             https://sergiocontrino.github.io http://www.intermine.org
>
>
>
>

_______________________________________________
dev mailing list
[hidden email]
http://mail.intermine.org/cgi-bin/mailman/listinfo/dev