You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@tomcat.apache.org by Mark Maigatter <Ma...@Scene7.com> on 2004/08/31 18:20:31 UTC
Java process growth under Linux...leak?
We have a Tomcat 5.0.25 based web site for uploading images and assorted
files and managing them.
We have found that the Java process that Tomcat is running under is
gradually growing when repetitively processing files uploaded and stripped
out of the form submissions by the Apache FileUpload component. All signs
point to a memory leak?
Upon the submission of about 500 files we had a 31MB growth in the size of
the java process using "top".
However, the Sun Java jvmstat shows that the Java heap is staying relatively
constant. The "-gc" numbers fluctuate in a manner that shows reasonable
garbage collection activity and the total used across the s0/s1/eden/old
stay with the range for the initial numbers.
My question is what would you recommend to isolate the process growth?
Is there a way within Java to see the underlying process growth to help
isolate it in the processing cycle?
--mark
Re: Java process growth under Linux...leak?
Posted by Wade Chandler <wc...@redesetgrow.com>.
Wade Chandler wrote:
> Mark Maigatter wrote:
>
>> We have a Tomcat 5.0.25 based web site for uploading images and assorted
>> files and managing them.
>>
>> We have found that the Java process that Tomcat is running under is
>> gradually growing when repetitively processing files uploaded and
>> stripped
>> out of the form submissions by the Apache FileUpload component. All
>> signs
>> point to a memory leak?
>>
>> Upon the submission of about 500 files we had a 31MB growth in the
>> size of
>> the java process using "top".
>> However, the Sun Java jvmstat shows that the Java heap is staying
>> relatively
>> constant. The "-gc" numbers fluctuate in a manner that shows reasonable
>> garbage collection activity and the total used across the s0/s1/eden/old
>> stay with the range for the initial numbers.
>>
>> My question is what would you recommend to isolate the process growth?
>>
>> Is there a way within Java to see the underlying process growth to help
>> isolate it in the processing cycle?
>>
>> --mark
>>
>
> How large were the files you were uploading? I mean you just said you
> uploaded 500 files. You should expect to see process memory growth. The
> JVM has it's own object heap where it manages it's "internal" memory.
> Then there is the process and it's memory which is a C heap.
>
> I see that you uploaded 500 files. Were these one right after the
> other, or were they close to simultaneous? Also, how are you uploading
> the files? Are you using some type of a parser? Are you using the
> commons file upload control? If the JVM reports good memory collection,
> then there is no memory leak in Tomcat. 31MB of growth for a process
> uploading 500 files shouldn't be that bad depending on how they were
> uploaded and file sizes.
>
> Think about it. If the files were between 100kb and 60kb then you have a
> total of 30mb to 50mb of memory just in that data alone not including
> your application and other buffers you may or may not be creating while
> uploading.
>
> For perforamnce reasons the VM isn't going to suddenly resize the heap
> as soon as it frees a group of java objects because as far as it knows
> you may come along and upload 50mb worth of data immediately after the
> first. This is a performance thing. Resizing the heap takes time and
> cpu resources and affects performance. The VM will reuse this memory
> over and over again.
>
> I would look at any loops I might have reading from the stream. Do you
> create a bunch of small byte array's while uploading the files? Maybe
> you could increase the buffer size, be sure to null them out after you
> perform a read to tell them VM you are done with the variable now (for
> when the vm collects), and then see if that affects the memory growth.
> This should help speed the file upload a bit and get rid of some buffers
> in loops a little quicker if you aren't nulling the array. If however
> you are simultaneously uploading much of this data, then a 31mb spike in
> memory usage shouldn't be a suprise no matter what.
>
> Basically you can impose limits on the VM and you can use switches to do
> this. You can also devote more memory to eden or survivor objects so
> that the VM can make better use of the most commonly used memory. You
> can find more info on this topic and others at this url:
> http://java.sun.com/docs/performance/
> Many docs. One for you might be:
> http://java.sun.com/docs/hotspot/VMOptions.html
> scroll down to the bottom and check out options:
> -XX:NewRatio
> -XX:NewSize
> -XX:SurvivorRatio
>
> Basically the defaults for the -server VM are to allow the best
> performance for a multi user multi threaded application such as tomcat.
> So, unless you are running out of memory or you need to cripple the app
> servers performance by limiting it's growth because you have a bunch of
> other applications running on the same server, then I suggest sticking
> with the defaults.
>
> Wade
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: tomcat-user-help@jakarta.apache.org
>
>
>
Just noticed you wrote about the upload component. I haven't looked at
that code's parse to see how it is handling reading bytes from the
stream, but I'm sure there have been many eyes on it. I would say this
is probably just a memory usage over time issue (short period of time).
Use those links I gave you and play with the memory switches a bit to
see if you can get some kind of a situation that works best for you.
The switches:
-XX:MaxHeapFreeRatio
and
-XX:MinHeapFreeRatio
Can make the vm resize the heap differently, but you will want to be
careful. You may make things very slow by messing with the defaults for
those switches.
Wade
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org
Re: Java process growth under Linux...leak?
Posted by Wade Chandler <wc...@redesetgrow.com>.
Mark Maigatter wrote:
> We have a Tomcat 5.0.25 based web site for uploading images and assorted
> files and managing them.
>
> We have found that the Java process that Tomcat is running under is
> gradually growing when repetitively processing files uploaded and stripped
> out of the form submissions by the Apache FileUpload component. All signs
> point to a memory leak?
>
> Upon the submission of about 500 files we had a 31MB growth in the size of
> the java process using "top".
>
> However, the Sun Java jvmstat shows that the Java heap is staying relatively
> constant. The "-gc" numbers fluctuate in a manner that shows reasonable
> garbage collection activity and the total used across the s0/s1/eden/old
> stay with the range for the initial numbers.
>
> My question is what would you recommend to isolate the process growth?
>
> Is there a way within Java to see the underlying process growth to help
> isolate it in the processing cycle?
>
> --mark
>
How large were the files you were uploading? I mean you just said you
uploaded 500 files. You should expect to see process memory growth.
The JVM has it's own object heap where it manages it's "internal"
memory. Then there is the process and it's memory which is a C heap.
I see that you uploaded 500 files. Were these one right after the
other, or were they close to simultaneous? Also, how are you uploading
the files? Are you using some type of a parser? Are you using the
commons file upload control? If the JVM reports good memory collection,
then there is no memory leak in Tomcat. 31MB of growth for a process
uploading 500 files shouldn't be that bad depending on how they were
uploaded and file sizes.
Think about it. If the files were between 100kb and 60kb then you have a
total of 30mb to 50mb of memory just in that data alone not including
your application and other buffers you may or may not be creating while
uploading.
For perforamnce reasons the VM isn't going to suddenly resize the heap
as soon as it frees a group of java objects because as far as it knows
you may come along and upload 50mb worth of data immediately after the
first. This is a performance thing. Resizing the heap takes time and
cpu resources and affects performance. The VM will reuse this memory
over and over again.
I would look at any loops I might have reading from the stream. Do you
create a bunch of small byte array's while uploading the files? Maybe
you could increase the buffer size, be sure to null them out after you
perform a read to tell them VM you are done with the variable now (for
when the vm collects), and then see if that affects the memory growth.
This should help speed the file upload a bit and get rid of some
buffers in loops a little quicker if you aren't nulling the array. If
however you are simultaneously uploading much of this data, then a 31mb
spike in memory usage shouldn't be a suprise no matter what.
Basically you can impose limits on the VM and you can use switches to do
this. You can also devote more memory to eden or survivor objects so
that the VM can make better use of the most commonly used memory. You
can find more info on this topic and others at this url:
http://java.sun.com/docs/performance/
Many docs. One for you might be:
http://java.sun.com/docs/hotspot/VMOptions.html
scroll down to the bottom and check out options:
-XX:NewRatio
-XX:NewSize
-XX:SurvivorRatio
Basically the defaults for the -server VM are to allow the best
performance for a multi user multi threaded application such as tomcat.
So, unless you are running out of memory or you need to cripple the
app servers performance by limiting it's growth because you have a bunch
of other applications running on the same server, then I suggest
sticking with the defaults.
Wade
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org