Java serialization is slow in Solaris 10

Hi,
I'm think to migrate from Linux to Solaris 10, because of the better support for Java any many other things, but in the first test the java web application, we get a too slower response in Solaris, with a time of 03 minutes and 12 seconds for one transaction, but this same transaction using Linux on same hardware and database the transaction concludes in 4 seconds, I think that this maybe a misconfiguration on Solaris.
In this application, all transaction are via JavaBean serialization over HTTP using an Applet that serializes/deserializes objects to a Servlet, seeing the tcp trace of http requests, after the server receives this requests, we get a pattern of 15 seconds to reply for every request.
Then, maybe a mistake of TCP configuration in Solaris? or Java Applet Serialization to Servlet is slow in Solaris?
Exists a best way to install Solaris for better performance with Java?
Thanks for any help or ideas
Cl�vis

I obtained a T1000 on the 60 day try and buy with solaris 10. I've installed my java application on it. On start up the first thing the application does is de-serialize a huge java object.
On my laptop (winXP single drive) this takes about 40 seconds.
On the T1000 (single drive c0t0d0 <SUN80G cyl 65533 alt 2 hd 16 sec 149) it takes about 12 minutes. The application is identical.
The java version on my laptop is 1.5.0_6
The java version on my T1000 is 1.5.0_10
I've doubled check to make sure the read and write cache is enabled on the disk.
This is a major problem for me. Any ideas? The T1000 is doing nothing else. It is 100% idle before I start the java application.
My application runs loads of threads and so I was hoping to get excellent performance which I haven't tested yet. However with a start up time of 12 minutes this is a no go straight away.
I understand it is more to do with the disc and solaris 10, but can anyone explain to me why my laptop should be so much faster?>

Similar Messages

Application runs slower on Solaris

We have developed an Application in Java (Servlets, JSP & Core Java classes including thread).
My Application runs fast on Windows, but its terribly slow on Solaris. When the application is run on Solaris the CPU usage is very less around 3 to 5 % only and there is no major io bound operation also. Please comment on this behaviour when am running the application on Solaris.
For eg: A while loop in my program took around 16 msecs to execute in Windows machine, but the same while loop is taking 250 msecs to run in solaris.
Note: The solaris machine is very powerful than the desktop windows machine.
Thanks & Regards,
Salin

Which version of Solaris are you running and what threading library is currently in use? Did you use some profiling application to arrive at the benchmark results you obtained or some other means were used? Does the code within the while loop call any classes making use of Java threads and synchronization mechanisms?

Need a lot of custom JComponents, but Java just too 'slow', other Solution?

Hi folks,
first of all i'll try to explain my problem (although my english is quite bad)
We're programming a tool for visualising graph algorithms (df search, strong connected components, transitive reduction etc.) For that, we got a drawing area (derived from JLayeredPane) which contains the Nodes and the Edges of a graph (for all that don't know what a graphs is, just imagine these nodes as numbered circles, and the edges as directed lines from one node to another).
Both parts (Nodes and Edges) are derived from JComponent, because a) need MouseEvents on these Components and b) we need the ability to add them to a container (like JLayeredPane)
An absolute requirement is that you can directly click on the edges (to mark them, or delete them etc)
Another point which needs to be mentioned, the edges maintain a (invisible) polygon which lies around their line, this polygon is sensitive for mouseEvents.
Now there are two main problems:
* all these edges have (due to JComponent) (possibly) huge invisible rectangles (the only thing that really matters is the line from one corner of this rect. to the other) ...so if i got lets say a graph with 10.000 edges, there are a lot of edges that overlap (nearly 10.000 :) ) ... but only the Edge-object on top receives the MouseEvent (but perhaps the user wanted to cklick a edge that lies below the top-edge!) .. at the moment, the program looks at all edges below that which received the MouseEvent and checks wether the 'mouse'click' hits a polygon, If it does, the top edge dispatches the MouseEvent to that edge. (furthermore, due to the depth-order in a Layer of a JLayeredPane, we only need to look at all edges that are really below the egde that received the mouseEvent)
This all works great for 'small' numbers of edges (approximately 1000-2000) but with a rising number of edges this attempt gets slower and slower (it takes 1-3 seconds to evaluate a single mouseclick on a Sun Ultra 5/333)
* the bigger problem is, that if i got 10k+ JComponents in a Container Java gets really slow, just adding these 10k JComps. (to the JLayeredPane) takes some minutes?! Furthermore java consumes up to 200+ MB main memory in this situation.
One solution we are currently thinking about is to implement these edges as 'pure graphics'-objects (for example the Container just draws all edges).. this probably solves problem # 2 but # 1 gets even worse (no depth-order, some more coordinates hassle)
So, i hope someone can give me a hint for this problem or someone has had a similar problem before and has a good solution? (if the problem isnt clear, just ask)

I worked on an application once that had a similar structure as yours, but we were only scaling up to about 1000-2000 components. The problem that we identified (and that you have, no doubt, also identified) is that you are searching all of the child components on every hit test/overlap test. What we did was to subdivide out outermost container so that it had a couple of dozen children, each of which then parented some of the original child components: this cut our search space for every hit test dramatically. We had the advantage of having collections of child componenents that we knew would not overlap so our first level containers did not overlap -- you may need your first level containers to overlap, but even with this, you'll still be able to much more quickly identify components in the immediate region that should be checked individuall.

Question concerning java Serialization of a complex internal field variable.

Not everything in the J2SE implements serializable by default. Say BufferedImage.
I have learned from an online artical that absolutely all fields to be serialized must
implement Serializable, including internal ("global") class fields,
a la
http://javarevisited.blogspot.com.au/2011/04/top-10-java-serialization-interview.html
(point 5).
However, for my purposes, I cannot re-implement (extend) and recompile every and any
java class I would want to serialize, eg. anything which is a "complex" sub field of
a conceptual class I do want to serialize, which doesn't occur on the java 1.6 list
of default serializable classes:
a la
http://docs.oracle.com/javase/6/docs/api/java/io/Serializable.html
and
"If member variables of a serializable object reference to a non-serializable object, the code will compile but a RumtimeException will be thrown."
Understanding there are implications for the java security model, is there anything to be done for these non-serializable-implementing class fields?
Is there a Serialization add on kit for Java or something?

Indeed, however, with a distinct lack of success with the instrumentation api, my request is dualfold now.
I do understand the restrictions explained here concerning more complex classes and the Serializable API.
I have found that there is an input line that needs to be submitted when running an Inustrumentation, premain Agent class.
However, I would have to avoid that by doing System.setProperty(a,b); appropriately.
Why won't the following just work, without any further supplied input parameters?
   import static java.lang.System.out;
   import java.io.PrintStream;
   import java.lang.instrument.Instrumentation;
   public class InstrumentationAgent
      private static volatile Instrumentation globalInstrumentation;
      public static void premain(String paramString, Instrumentation paramInstrumentation)
         System.out.println("premain...");
         globalInstrumentation = paramInstrumentation;
      public static void agentmain(String paramString, Instrumentation paramInstrumentation)
         System.out.println("agentmain...");
         globalInstrumentation = paramInstrumentation;
      public static void main(String ... args)
         out.println("!");
         String data = new String("data");
         long size = getObjectSize(data);
         out.println(size);
      public static long getObjectSize(Object paramObject)
         if (globalInstrumentation == null)
            throw new IllegalStateException("Agent not initialized.");
         return globalInstrumentation.getObjectSize(paramObject);
I am still curious as to how the DataFlavor approach I have outlined above. Does it accomplish a shallow or deep conversion? Are
there other limitations which restrict what's put to the underlying ByteOutputStream?
Why won't my agent class execute though, regardless?

My java is extremely slow in yosemite

my java is extremely slow in Yosemite. i found existing comments on uninstall java and reinstall. I did reinstalled java couple of times, but not helping.
when i simply enter java, java -version, javac in the terminal, its taking **** lot of time to respond. so not able to run even single simple java program. but other softwares are working fine. i am having python and running python code is pretty fast.
how to resolve this?

Apple doesn’t routinely monitor the discussions. These are mostly user to user discussions.
Send Apple feedback. They won't answer, but at least will know there is a problem. If enough people send feedback, it may get the problem solved sooner.
Feedback
Or you can use your Apple ID to register with this site and go the Apple BugReporter. Supposedly you will get an answer if you submit feedback.
Feedback via Apple Developer

Java is very slow on terminal servers

Hi, we have several terminal servers for clients. But running Java (JRE) on a bank website, is very slow to load the Java plugin. So the whole login takes around 1-2mins to complete.
We run Windows Server 2008 R2 which is virtualized with VMWare, we run the latest Java client v7 Update 17. Java has always been slow on these servers, is there a way to teak it somewho, so it loads faster?

WoOber wrote:
Its me who have setup the whole server, but users experience really that Java starts really slow when they try to login on their bank on the web.And how is that login done? In an applet? Because then the performance issue is on their own computer in their own private installations of browsers and Java runtimes, not your terminal. Applets tend to load 'slowly', especially on older versions of Java. The feeling of slowness tends to be because people compare the loading time to Flash, a completely different beast.

Java 1.6.0 on Solaris 10?

I'm writing an application that uses Java 1.6.x, so I've been trying it out on various operating systems...
Windows : peice of cake, installs and runs first time.
Linux : peice of cake, installs and runs first time.
Mac OS : need to download java beta 1.6 (or 6.0), but it does work.
Solaris : Oh my god...how the hell can I get this to work?
I can't believe how hard it is to run a java 1.6 applications on Solaris 10?
hang on a minute, doesn't Sun make Solaris?...and also make Java?
So, whats the trick, how do I install java 1.6 on Sun's Solaris 10 (x86) ?
Thanks.
p.s. Sorry if I'm being stupid, but it doesn't make me want to run Solaris.

I ordered a new machine recently and had to load Java 6 manually. I was looking for a Sun package that would do this automatically and couldn't find one.
In brief, these are the steps I used:
1. Download the self extracting version of the install. In my case, it was sparc:
jdk-6u2-solaris-sparc.sh
2. As root, cd to /usr/jdk and run the script.
Myserver# cd /usr/jdk/instances
Myserver# sh jdk-6u2-solaris-sparc.sh
3. This should give you a new JDK directory like this:
Myserver# ls -l
total 4
drwxr-xr-x 9 root bin 512 May 17 16:43 jdk1.5.0
drwxr-xr-x 9 root bin 512 Aug 3 09:00 jdk1.6.0_02
4. Now you need to point all the links in the /usr/jdk at the new instances/jdk1.6.0_02 directory.
Myserver# cd /usr/jdk
Myserver# ls -l
total 12
drwxr-xr-x 4 root bin 512 Aug 3 09:00 instances
lrwxrwxrwx 1 root other 7 May 16 14:59 j2sdk1.4.2_12 -> ../j2se
lrwxrwxrwx 1 root other 18 May 17 16:44 jdk1.5.0_11 -> instances/jdk1.5.0
lrwxrwxrwx 1 root root 21 Aug 3 09:00 jdk1.6.0_02 -> instances/jdk1.6.0_02
lrwxrwxrwx 1 root root 11 Aug 3 09:01 latest -> jdk1.6.0_02
drwxr-xr-x 7 root bin 512 May 16 15:19 packages
Does make me wonder if I've missed something or done it the hard way.

Using JAXB 2.0 as the Java Serializer for WSIF bindings

Hello All:
I am trying to invoke a Java Class from a BPEL process (using native WSIF Java Binding) using JAXB 2.0 as the Java Serializer. However it seems that Oracle already has an older verion of JAXB on the classpath which results in an exception during the serialization (could not find jaxb.properties in the package ...). Can you guys tell me how to pick up the newer version of the JAXB library. I have added the required jar files as a shared library. How can the classloading behaviour be altered to accomodate this?
Does 10.1.3.4 take care of this? I have 10.1.3.3 currently

Hi,
Did you ever solve this issue? I'm having the same problem with toplink and hibernate jars....
Regards,
Zaloa

Customizing ActionScript - Java serialization?

Hi,
I wonder how to customize what ActionScript class is serialized to what Java object, overriding BlazeDS's defaults.
I have a Java class where I would like an ActionScript ArrayCollection to serialize to a custom class.
To illustrate, normally I might have a class on the server like this:
public MyClass {
public void updateRecords(java.util.List[Widget]) {...}
When this is invoked from a remote Flex client, passing an ArrayCollection as the method parameter, BlazeDS serializes that ArrayCollection into a Java List, and everything works fine.
Suppose, however, that I have the above method signature defined in the following way:
public MyClass {
public void updateRecords(MyCustomCollection[Widget]) {... }
In that situation, I'd like to be able to specify a custom serializer that takes an ActionScript ArrayCollection and serializes it into MyCustomCollection. In fact, what I'd really like to do is to use this serializer every time a Java class' interface contains MyCustomCollection.
I looked into flex.messaging.io.TypeMarshaller, but couldn't quite figure out how to use it for that purpose.
Thanks,
-- Frank

>
How Java Serialization works internally which algorithm it uses to serialize/deserailaize the objects?
>
See the Java Object Serialization Specification
http://docs.oracle.com/javase/6/docs/platform/serialization/spec/protocol.html
And are you aware that you have never marked a single one of your 119 previous questions as answered?
Total Questions: 119 (119 unresolved)
Are all of those questions really unanswered? Or have you just forgotten to mark them ANSWERED.
Please begin revisiting your large number of previous posts and marking them ANSWERED as appropriate. This is important for several reasons. First because forum volunteers need to focus on unanswered questions - it wastes their time to read thru threads and respond to posts that have actually already been answered since the original poster isn't even going to pay attention to any new replies. That time could have been better spent helping users that have REAL unanswered questions on the forum.
Second, when someone like yourself has large numbers of unanswered questions it basically sends a signal to the volunteers that the person posting the question may not even be reading or responding to the replies the volunteers may make. This leads some volunteers to decide to not even bother responding - why should they spend their time trying to help someone who won't even acknowledge the help that they get?
Third, other people that have the same problem you had find your posts but never see them answered. So they never know if the information provided by the volunteers solved your problem or not. When you mark your questions answered it helps others with the same problem.
Thanks.

Why is Java startup so slow?

Does anyone know why Java startup is so slow? Compared to other interpreted languages like Javascript and Perl, they run much faster. And can someone get technical with me? (But without pointing me to the JVM specification, which I don't understand.) It use to take longer for me to load Mozilla0.9 than a Java applet, but now Mozilla1.3 loads faster than a Java applet.

"Connector" is a rather ill-defined way of describing
it, which is why I chose the word -- to easily deflect
further inquiries to quantify. ;>
In a slightly more specific way, on start-up JVM needs
to (not necessarily in order, 'cause I don't know the
exact order) at the very minimum:
1. Allocate memory.
2. Query OS for a minimal set of needed facilities.
3. Allocate memory. Allocate allocated memory
(internally, for use by itself and the application).
4. Map the OS facilities to Java facilities. (This is
what I meant by "connectors").
5. Allocate memory. Allocate allocated memory.
6. Set up the internal control -- security managers,
thread scheduler, gc, classloaders, calling stacks,
virtual lookup tables, etcetc.
Some of the above don't apply to applets.
Thanks for the tip on 1.4.2beta. I'll give it a try. I
assume you meant startup speed?All this is almost instant you realize. The real time it takes to start your program is that it takes much longer for Java classes to be loaded.
The startup time is proportional to the number of classes you need to load to get your program running. If you only are writing a simple, console based java program the startup time will be much less then if you load a Swing application. Swing by itself is a huge number of classes, and then you have to load your classes. So it can take awhile.
All the things you mention, every program has to do and the code that the JVM uses must be written in C or some other systems language.
One reason at least on my machine that java startup is slow is that I have a slow hard drive. On my home machine, java is much faster around twice as fast to startup, but the processor and ram are not twice as good. I think that it's because the hard drive is faster. Because most of the time it's waiting for data from the disk and the Java enviroment has the disadvantage that it's very large on disk.

Java 2D very slow with Windows XP

Hi all!
I'm writing an isometric game engine using Java 2D API, in full-screen exclusive mode and using all the advantages of the new VolatileImage class. The engine is almost finished, I'm testing it putting some animated players on screen and showing the frame rate of the engine. Everything went right, frame rate was very high in Linux and Windows 98/Me. But when I try to run the engine on Windows XP I can only get 5 or 6 fps!!!! And I don't know why!! It seems that XP makes Java run very slow... at least when using Java2D...
Please, anyone knows why the performance is reduced??? Is there a new JRE release for Windows XP?? I'm using JDK 1.4.1.
Thanks for your help.

If you're using the Fullscreen API, then you're using BufferStrategy, thus you shouldn't use VolatileImages, it does this for you in the background. You shouldn't create any buffers, simply use the one provided. I'm assuming your problem probably lies in transparency. I've made an isometric engine using the fullscreen API that has run with complete 60fps page flipping. I would assume your bottle-neck lies in:
Transparencys (These are very slow, there is a way to make a bitmask, much like you'd do in DirectX)
Dynamic Memory allocation/Alpha rendering in your animation loop. (You need to allocate ALL of the colors you're going to use, and avoid using an alpha component if you want to avoid the VM getting bogged down by the garbage collector because you're allocating 30 Color objects a second).
I made a Color object that was some what hacked together that has methods to allow you to access it's values for doing this kind of thing, allocating your scheme before hand is an easier approach.
The method that performs the page flipping is a BLOCKING method, so you should orient your drawing before it, and not assume it's simply placing a request, like repaint does.
-Jason Thomas.

Default to Java Serialization in case Pof Serialization not defined

Is this possible to do, i.e. essentially if for a certain class Pof serialization is not defined, use the Java serialization instead?
Or to turn it around, is it possible to define pof serialization only for certain classes in a distributed cache and use Java serialization for the rest?

Hi,
the problem for this is that Java serialization is not aware of POF (or for that matter even ExternalizableLite), so if you have a Java-serialized class which has a member which is supposed to be POF-serializable, it in fact will not be serialized with POF, because Java serialization will not delegate to POF.
So it is very hard to mix the two together. You can do it for top-level objects by providing a special PofSerializer instance for the non-POF class which serializes to byte array and you write the byte array as a POF attribute, but it is not possible for POF-aware objects contained within a non-POF aware object to be POF serialized.
Also, if you attempt to do this, then you can kiss goodbye to platform independence. You must use Java on both ends and have all the libraries which the classes used in the state want to pull in.
Best regards,
Robert

Java Webservice client slow compared to dot net client

we have DotNet webservice which is being called by two clients , one a dotnet client and other a javaclient
The problem is java client is taking longer time in processing the results when compared to dotnet client and affecting the performance greatly , i'm using axis in java
Is there any know issue with axis regarding my situation here.
also the same java client when run on linux platform still longer time to process.
please provide me some pointer here

I would test the general performance difference between the 2 platforms on the same machine.
I have the feeling that Java is generally slower on windows than .NET
Microsoft has their secrets after all.

Java profiler for WLS6 on Solaris.

Hi,
Is there any profiler that supports the Sun JDK 1.3.0 on Solaris, as
OptimizeIt doesn't even start and JProbe hangs when trying to profile
the server.
Thanks,
Deyan
[dejan_bektchiev.vcf]

To make it work a little better (but 100 times slower ;-) turn off hotspot
and jit!!!
Peace,
Cameron Purdy
Tangosol, Inc.
http://www.tangosol.com
+1.617.623.5782
WebLogic Consulting Available
"Deyan D. Bektchiev" <[email protected]> wrote in message
news:[email protected]..
Yes,
It is more stable than the others -- doesn't crash sometimes for 10minutes
but it's not enough for me to collect performance data but I am still
trying so eventually I might be able to gather some.
I keep getting
# HotSpot Virtual Machine Error, Unexpected Signal 11
# Please report this error at
# http://java.sun.com/cgi-bin/bugreport.cgi
# Error ID: 4F533F534F4C415249530E435050079A 01
# Problematic Thread: prio=5 tid=0x2a3328 nid=0x8 runnable
but since without the profiler settings I do not get the problem I am
unable to say what happens and the behaviour is not consistent --sometimes
the JVM crashes at WLS startup sometimes later.
Thanks,
Deyan
Cameron Purdy wrote:
Have you tried the JDK's profiler?
Cameron Purdy
Tangosol, Inc.
http://www.tangosol.com
+1.617.623.5782
WebLogic Consulting Available
"Deyan D. Bektchiev" <[email protected]> wrote in message
news:[email protected]..
Hi,
Is there any profiler that supports the Sun JDK 1.3.0 on Solaris, as
OptimizeIt doesn't even start and JProbe hangs when trying to profile
the server.
Thanks,
Deyan

Sun JDK 1.3 slower on Solaris 8 than Linux 6.2

I'm doing platform testing to determine which OS will run my application the fastest. I
have a RH Linux 6.2 box with dual p3-700's /512 Mb RAM running WLS 5.1 sp 5 with
Sun's JDK 1.3 and a Sun E-250 running Solaris 8, with dual Sparc 400's / 512 Mb
RAM running WLS 5.1 sp5 with Sun's JDK 1.3 Amazingly, the Linux box runs about
2x to 3x faster under load than the Sun box. Based on the hardware being used, one
would assume the Sun box would be AT LEAST as fast if not FASTER than the Linux
box. Surely, Sun didn't write a JDK for Linux that outperforms Solaris. There are alot
of different features/switches to the 1.3 JDK, I tried using different combinations and
so far came to the conclusion that the JVM should be invoked with the following:
- server -XX:+UseThreadPriorities -Xms256m -Xmx384m

If you are just looking for raw performance, you will do better on x86.
Even at the same mhz, you will typically get better performance from an x86
processor (e.g. P3). If you are looking to save money up front, obviously
you can't beat commodity x86 hardware running Linux. There are those who
believe that you save a lot of money (and headache) over the long term by
going with Sun Sparc running Solaris.
Cameron Purdy
Tangosol, Inc.
http://www.tangosol.com
+1.617.623.5782
WebLogic Consulting Available
"Chris" <[email protected]> wrote in message news:[email protected]...
>
Well the question I was asking was what kind of analysis to do.
I'm not sure what else to tweak, configure, or manipulate. The
app doesn't hit a db in testing, rather we have it reading from a
rowset locally on disk to eliminate any latency associated with db access.It is highly CPU bound, the piece of the app in question processes data
through some complex algorithms to provide data back to the web server. Do
you think the SPARC 400's are really 2x slower than the P3-700's? If that's
the case, then
it would be a hard sell to run a java app on a Sun box at 5x the cost ofrunning it on Linux.
>
Thanks,
-Chris
Rob Woollen <[email protected]> wrote:
You'd have to do some more analysis to figure out why the Linux box is
faster.
What is your application doing under load? Are you hitting a database?
Are you cpu-bound, io-bound etc?
FWIW, if you were cpu-bound, a p3-700 is quite a bit faster than a
Sparc-400. However, web applications are rarely cpu-bound.
-- Rob
Chris Cosie wrote:
I'm doing platform testing to determine which OS will run my
application the fastest. I
have a RH Linux 6.2 box with dual p3-700's /512 Mb RAM running WLS 5.1sp 5 with
Sun's JDK 1.3 and a Sun E-250 running Solaris 8, with dual Sparc 400's/ 512 Mb
RAM running WLS 5.1 sp5 with Sun's JDK 1.3 Amazingly, the Linux boxruns about
2x to 3x faster under load than the Sun box. Based on the hardwarebeing used, one
would assume the Sun box would be AT LEAST as fast if not FASTER thanthe Linux
box. Surely, Sun didn't write a JDK for Linux that outperforms Solaris.There are alot
of different features/switches to the 1.3 JDK, I tried using differentcombinations and
so far came to the conclusion that the JVM should be invoked with thefollowing:
- server -XX:+UseThreadPriorities -Xms256m -Xmx384m

Java serialization is slow in Solaris 10

Similar Messages

Maybe you are looking for