What I'm hoping to see is an open solution that I/everyone can contribute to so we have a good open standards compliant PIM software. Currently all of the folks that write PIM software for cell phones have created their own flavours of iCalendar and vcard. They improve their implementations with each new cell phone and each revision of each new cell phone, but quite a lot of limitations and problems still exist. It's hard to support moving targets like this and the targets have been moving slowly for over 4 years.
Personally (as the author of a SyncML solution and someone who has to support n phones/revisions across the globe) I haven't seen any cell phone devices use anything but the older deprecated VCARD standard. I also don't remember (very tired though) seeing any devices that support the newer IETF iCalendar standard. I believe an open solution would be able to get out an updated implementation of the newer standards and enable a much higher level of functionality and interoperability. I believe a better PIM foundation will enable more interesting and useful applications and services to be built.
It's not uncommon for a product to be released initially for free in order to get as many people as possible using the system providing feedback. I use the feedback to build the features folks want, which in turn causes more people to subscribe, which causes more feedback, etc. Free simply means a much stronger and diverse and quickly growing community.
ScheduleWorld uses open standards (iCalendar) and your data is never held hostage. Simply switch to any other product that uses iCalendar at any time you wish. The web site states this clearly. The question of wether or not I may charge for the service in the future does not change this.
I believe it's inevitable that ScheduleWorld will move to an open source license as well. Though technically all other competing products are far behind ScheduleWorld it's only a matter of time before the open source solutions catch up and then there will be little point in being closed.
I use animtree for the tree on the left that you say doesn't work in Safari. Perhaps my version of animtree is obsolete. Can anyone running Safari view the animtree home page demos properly?
http://dhtmlkitchen.com/scripts/animtree/demo/in de x.jsp
BTW, you'll find the ScheduleWorld GUI pleasant and snappy under OS/X. And as soon as Apple releases Java 1.5.0_01 (_01 is important) I should be able to have a mostly perfect OS/X look and feel for the OS/X folks.
If you do a google search on outlook interoperability you'll see my product ScheduleWorld as the first link behind Microsoft. I've spent tremendous resources over the years trying to adapt and work around all of the problems with Microsoft Outlook/Exchange and their implementation of open iCalendar (rfc 2445, 2446, 2447) standards.
If my company of one person can make a better implementation of an open standard (this does not include the GUI or the help) than Microsoft then I have to wonder exactly how important it is to Microsoft that they get it right and interoperate properly.
ScheduleWorld is free, works on Linux, Mac OS/X, Windows, Solaris. It is by far more standards compliant and interoperates really well with Exchange/Outlook and Notes. Check it out and see for yourself.
By and large there is no need to demand your users trust you with full write access to their home directory, their ethernet device, and more. Consider writing your software in the Java Web Start sandbox.
The Java Web Start sandbox environment may be a bit too limited for some applications, but it is secure and more applications are being written for it all the time. Sun is also improving it with every release. In this environment you don't have to trust the code, or the software vendor wrt manipulating your hard drive, network interfaces, keyboard, or even the clipboard.
"Whose brilliant idea was it to give corporations the same legal rights as an individual?"
You hit the nail on the head. Additionally disturbing: the documentary "The Corporation" http://www.thecorporation.tv/filmmak ers/ makes a strong case for defining a corporation as the perfect psychopath.
ScheduleWorld is a replacement for Outlook. We provide a freely available Exchange-like calendar service too.
All operations are asynchronous so it works offline. The advantages online are that tasks that that take time (like creating an event over the Internet) appear to happen in zero time and don't slow you down for a microsecond.
All of your working set (your data, and all other data you need) is cached locally. Everything is kept in sync via async background threads passing compressed diffs when necessary.
Your account can be active on several PCs at once. All of them kept in sync.
You can sync your data (Windows, OS/X, Linux, Solaris) to your cell phone via SyncML. You can sync your Palm, PPC, Outlook, Exchange, etc. hardware/software devices via SyncML too.
Worried about security? Here are two interesting facts: 1. ScheduleWorld runs in a Sandbox and can't snoop your keystrokes, read your hard drive or install spyware. Here's something new: you don't have to trust the application you are running with respect to spyware/viruses and other such nonsense because it simply _can't_ do any of those things.
2. All of your data is encrypted. The ScheduleWorld admins can't read it. For those wearing tinfoil hats: the FBI could subpoena your calendar/contacts but they can't read them either unless you give them your private keys.
ScheduleWorld raises the bar for Interent applications to a whole new level. It's the way software should be done.
I will debate logical arguments and answer any reasonable questions about any security/performance related issues.
Free, based on open standards, interoperates with Exchange/Outlook/iCal/Notes/etc... server hosted for you for free, encrypted todos/events, multi-machine sync (soon also SyncML), Java Web Start client to support OS/X, Linux, Solaris, Win32, and more...
Lots of interesting things could be done. Some problems: 1. how to resolve disputes when no central authority exists 2. how to prevent hackers from destroying the integrity of the system
I've always thought that DNS should be enhanced to provide the CA for the domain. This would let people be their own CA, and the CA install process would be automated. Also, a CA should only EVER be an authority for a specific domain.
I visited the site, and the commercial site too and it seems this is only simple replication with the master being a single point of failure. F.E.
1. update a row in the master 2. master replicates the update to multiple slaves 3. clients perform select operations against the slaves (nice load balancing opportunity) 4. the master crashes 5. No one can write until the master comes back online.
Here are the steps that seem to be missing: 6. the slaves elect a new master 7. if the old master comes back up it must realize a new master is present and become a slave. 8. clients using JDBC would need some mechanism of finding out what the new master is when an update/insert/delete fails.
Cheers.
Official Open[GL,AL] bindings for Java
on
OpenGL 1.5
·
· Score: 3, Informative
Sun announced full Open[GL,AL] support here:
http://games-core.dev.java.net/
Here's a great example of using OpenGL/OpenAL under Win32/Linux written in Java. (It uses the LWJGL - which is an OpenGL/OpenAL Java wrapper that uses nio).
2. The client is in Java and works on Linux, Win32, and the Mac.
3. It is very secure. An interesting and short description may be found here: http://www.scheduleworld.com/itsYourLife.ht ml
4. It uses Java Web Start to keep it up to date. Once you install it you will never have to install or update it again (happens automatically).
5. It uses the same format as Outlook/Lotus Notes and you can export your data at any time if you want to move to another product.
6. Outlook interop isn't quite ready yet, and the SOAP API isn't quite ready for public exposure yet but both will get there.
7. Corporate time sheet extensions (with full SR&ED - Scientific Research and Experimental Development - tax credit reporting) will be completed and ready for use in less than 2 weeks. This would be a "for pay" feature.
1. FUTEX - a fast user-mode mutex. In the likely (hopefully) usual case no context switch happens when you get a futex. This can be a big win under certain synchronization loads. Imagine grabbing a futex in 15 clks vs 1,500 - 6,000 clks (these numbers may be slightly off and will vary quite a bit between machines, glibc and linux kernel versions)
2. NPTL - the New Posix Threading Library - a great improvement in both Posix compliance, performance, latency, and in the maximum number of usable threads. Ingo (one of the authors) announced the creation and destruction of 100,000 threads in one second. This is available in RedHat 9. (I don't know if futexes are available in the RedHat 9 kernel yet).
3. Graphics - Hey, I'm a Linux/Java fan and I'd like to be able to agree with you here but no. Doing a search in Sun's Java bug list will show up a lot of bugs in Linux graphics performance. In fact, check out http://www.javagaming.org/ and browse some of the posts there about it. There are tons of benchmarks of actual games that get 1/10 of the video throughput under Linux vs Win32. Sorry. (BTW, it seems the Blackdown folks are claiming an 80% 2D speedup in their latest port. I haven't tried it yet but it may help).
4. About ALSA - it rocks! What I meant to say was that perhaps RedHat could help the ALSA project support more cards and remove the completely stunned ARTS/ESD/etc... hacks and replace them with the proper ALSA DMIX plugin. This would allow multiple/dev/dsp open()s to work on cards that don't provide hardware mixing. This means you can play Java sounds and run KDE/Gnome sounds at the same time. (JDK 1.4.2-beta already has ALSA and DMIX support BTW - comments welcome on how well it works).
As others have mentioned the Java source is already freely available. As someone who spends most of their time in Java I can already fix bugs if I need to - having a separate license (GPL) is meaningless to me.
It would be a much better use of time and talent to make Java work better on Linux wrt: 1. FUTEX support 2. NPTL threads 3. full screen and 2D graphics are horribly slow under Linux because for some reason Sun doesn't seem to use MIT-SHM, or their pixmap caching code is doing the wrong thing... 4. Why not even spend time helping with Sun's ALSA port for Java 1.4.2.
Heck, I'm sure the Blackdown team would have dozens of ideas on how to improve the existing code base.
Rewriting from scratch? Is working together so hard? It would be such a shame to have great Linux coders work to build something that didn't work perfectly and was never used.
They do seem off. Even a 66MHz FSB wouldn't account for it. Here's an interesting post by Linus that mentions the old "int 0x80/iret" was 1761 and the new sysenter/sysexit is 641 (on a P4):
http://www.ussg.iu.edu/hypermail/linux/kernel/0212.2/0261.html
If you use the assembly language instruction RDTSC right before and after getpid() to determine how many clock cycles a simple (maybe the simplest) context switch can take you will find some interesting results:
(RDTSC gives a correct result +- 20clks) 400MHz PII: 44,000 PIV: 18,000 266MHz Athlon 1.3GHz: 6000
These numbers are from memory from my tests several years ago. I have seem posts on lkml that glibc/Linux may have the PIV case down to about 9,000 ckls but I haven't tested it. However, I am fairly certain the PII case is correct - and the 66MHz FSB was mostly to blame for the bad numbers.
Let's take Apache using on average 10 (guess) system calls for 10 x 44,000 = 440,000 clks for a 6-line index.html file. Apache would only be able to serve 909 index.html files/second using 100% of the CPU. (simplistic)
Using Tux, we would be able to serve 10,000 index.html files/second using 100% of the CPU.
In the real world, Tux is able to server 10,000/s on my 400MHz PII, and Apache is able to server over 2,000 so perhaps it doesn't use 10 System calls/page (I was guessing). In any case, the performance difference can obviously be worth while.
Generics is auto-casting mechanism and compile-time type checking - nothing else. If you feel otherwise let us all know why.
Note: I did not say Generics was bad, or that it wasn't an improvement, or that it affected performance.
I just want to be clear: I mentioned fastutil because I feel in many cases developers would prefer to use it over Generics for these reasons: 1. It allows you to use native data types in collections. 2. It provides more algorithms 3. When used with native data types, it is enormously faster, much more memory efficient, and saves the garbage collector from potentially large amounts of needless work.
Generics is an auto-casting mechanism that is useful to catch programming errors at compile time instead of run time. Please note 2 things:
the 'cast' still happens at runtime as the compiled bytecode is still the same.
there is a much faster alternative that uses native data types instead of objects (below) and catches errors at compile time.
As a pro Java developer, I want to use the native 'int' type in order to save memory, have less garbage collection, and perform better. Catching errors at compile time is helpful too. I think it is unreasonable for Sun not to include specializations for native data types. If I want to have an ArrayList of 10,000 ints I should be able to use 'int'.
The link on this page states up to 10x performance but I've seen it work up to 30x performance - and you can run the code below to see this for yourself. NOTE: 30 = 7272727/236966 where: 1. 7272727 = 2nd iteration of Int2IntHashMap! 2. 236966 = 15th iteration of HashMap (Hot spot had 12 more iterations to optimize)
http://fastutil.dsi.unimi.it/
package com.wss.utils.test; import java.util.*; import it.unimi.dsi.fastUtil.*; public class TestFastUtil {
public static void main(String[] args) throws Exception {
int count = 400000;
int timerCount = 20;
long start, end;
Integer tmp;// Normal HashMap
HashMap hashMap = new HashMap(count);
for (int timer=0; timer timerCount; ++timer) {
start = System.currentTimeMillis();
for (int i=0; i count; ++i) {
hashMap.put(new Integer(i), new Integer(i));
}
end = System.currentTimeMillis();
System.out.println("HashMap put(Integer, Integer) count:" + count +
", put/s:" + (count / ((float)(end-start) / (float)1000)));
start = System.currentTimeMillis();
for (int i=0; i count; ++i) {
Integer in = new Integer(i);
tmp = (Integer)hashMap.get(in);
if (!tmp.equals(in))
throw new Exception("failed equals()");
}
end = System.currentTimeMillis();
System.out.println("HashMap get(Integer) count:" + count +
", get/s:" + (count / ((float)(end-start) / (float)1000)));
}// FastUtil HashMap Int2Int
timerCount = 100;
Int2IntHashMap int2IntHM = new Int2IntHashMap(count);
int j;
for (int timer=0; timer timerCount; ++timer) {
start = System.currentTimeMillis();
for (int i=0; i count; ++i) {
int2IntHM.put(i, i);
}
end = System.currentTimeMillis();
System.out.println("Int2Int put(Integer, Integer) count:" + count +
", put/s:" + (count / ((float)(end-start) / (float)1000)));
start = System.currentTimeMillis();
for (int i=0; i count; ++i) {
j = int2IntHM.get(i);
if (i != j)
throw new Exception("Int2Int failed equals()");
}
end = System.currentTimeMillis();
System.out.println("Int2Int get(Integer) count:" + count +
", get/s:" + (count / ((float)(end-start) / (float)1000)));
}
}
}
It would seem to me that it's clear this patch is much more than just a non-executable stack patch. I think an important point is that while it isn't perfect, if this becomes a standard part of Linux then we can stand on the shoulders of this and tackle the next set of problems. Eventually we may get there.
If we wait until a perfect patch presents itself we will never get there. F.E. using this patch will bring to light all applications that require gcc trampolines. With time, perhaps all of the trampoline code will be replaced - several articles have been written on how to live without trampolines, perhaps this would be the push needed to change.
Slashdot Mirror
What I'm hoping to see is an open solution that I/everyone can contribute to so we have a good open standards compliant PIM software.
Currently all of the folks that write PIM software for cell phones have created their own flavours of iCalendar and vcard. They improve their implementations with each new cell phone and each revision of each new cell phone, but quite a lot of limitations and problems still exist. It's hard to support moving targets like this and the targets have been moving slowly for over 4 years.
Personally (as the author of a SyncML solution and someone who has to support n phones/revisions across the globe) I haven't seen any cell phone devices use anything but the older deprecated VCARD standard. I also don't remember (very tired though) seeing any devices that support the newer IETF iCalendar standard. I believe an open solution would be able to get out an updated implementation of the newer standards and enable a much higher level of functionality and interoperability. I believe a better PIM foundation will enable more interesting and useful applications and services to be built.
Cheers.
It's not uncommon for a product to be released initially for free in order to get as many people as possible using the system providing feedback. I use the feedback to build the features folks want, which in turn causes more people to subscribe, which causes more feedback, etc. Free simply means a much stronger and diverse and quickly growing community.
ScheduleWorld uses open standards (iCalendar) and your data is never held hostage. Simply switch to any other product that uses iCalendar at any time you wish. The web site states this clearly. The question of wether or not I may charge for the service in the future does not change this.
I believe it's inevitable that ScheduleWorld will move to an open source license as well. Though technically all other competing products are far behind ScheduleWorld it's only a matter of time before the open source solutions catch up and then there will be little point in being closed.
I use animtree for the tree on the left that you say doesn't work in Safari. Perhaps my version of animtree is obsolete. Can anyone running Safari view the animtree home page demos properly?
n de x.jsp
http://dhtmlkitchen.com/scripts/animtree/demo/i
BTW, you'll find the ScheduleWorld GUI pleasant and snappy under OS/X. And as soon as Apple releases Java 1.5.0_01 (_01 is important) I should be able to have a mostly perfect OS/X look and feel for the OS/X folks.
Cheers.
If you do a google search on outlook interoperability you'll see my product ScheduleWorld as the first link behind Microsoft. I've spent tremendous resources over the years trying to adapt and work around all of the problems with Microsoft Outlook/Exchange and their implementation of open iCalendar (rfc 2445, 2446, 2447) standards.
If my company of one person can make a better implementation of an open standard (this does not include the GUI or the help) than Microsoft then I have to wonder exactly how important it is to Microsoft that they get it right and interoperate properly.
ScheduleWorld is free, works on Linux, Mac OS/X, Windows, Solaris. It is by far more standards compliant and interoperates really well with Exchange/Outlook and Notes. Check it out and see for yourself.
By and large there is no need to demand your users trust you with full write access to their home directory, their ethernet device, and more. Consider writing your software in the Java Web Start sandbox.
There are techniques the Mozilla folks could have used to build software that does not require your trust. http://www.scheduleworld.com/itsYourLife.html
For more secure Java Web Start info: http://www.scheduleworld.com/itsYourLife.html
"Whose brilliant idea was it to give corporations the same legal rights as an individual?"
k ers/
You hit the nail on the head. Additionally disturbing: the documentary "The Corporation"
http://www.thecorporation.tv/filmma
makes a strong case for defining a corporation as the perfect psychopath.
ScheduleWorld is a replacement for Outlook.
We provide a freely available Exchange-like calendar service too.
All operations are asynchronous so it works offline. The advantages online are that tasks that that take time (like creating an event over the Internet) appear to happen in zero time and don't slow you down for a microsecond.
All of your working set (your data, and all other data you need) is cached locally. Everything is kept in sync via async background threads passing compressed diffs when necessary.
Your account can be active on several PCs at once. All of them kept in sync.
You can sync your data (Windows, OS/X, Linux, Solaris) to your cell phone via SyncML. You can sync your Palm, PPC, Outlook, Exchange, etc. hardware/software devices via SyncML too.
Worried about security? Here are two interesting facts:
1. ScheduleWorld runs in a Sandbox and can't snoop your keystrokes, read your hard drive or install spyware. Here's something new: you don't have to trust the application you are running with respect to spyware/viruses and other such nonsense because it simply _can't_ do any of those things.
2. All of your data is encrypted. The ScheduleWorld admins can't read it. For those wearing tinfoil hats: the FBI could subpoena your calendar/contacts but they can't read them either unless you give them your private keys.
ScheduleWorld raises the bar for Interent applications to a whole new level. It's the way software should be done.
I will debate logical arguments and answer any reasonable questions about any security/performance related issues.
"Because the web-based interface of OPEN-XCHANGE? runs on all major browsers".
For those of you who don't want to wait for each and every mouse click check out ScheduleWorld.
Free, based on open standards, interoperates with Exchange/Outlook/iCal/Notes/etc... server hosted for you for free, encrypted todos/events, multi-machine sync (soon also SyncML), Java Web Start client to support OS/X, Linux, Solaris, Win32, and more...
Lots of interesting things could be done.
Some problems:
1. how to resolve disputes when no central authority exists
2. how to prevent hackers from destroying the integrity of the system
Thoughts?
I've always thought that DNS should be enhanced to provide the CA for the domain. This would let people be their own CA, and the CA install process would be automated. Also, a CA should only EVER be an authority for a specific domain.
Please let me know if I'm wrong...
I visited the site, and the commercial site too and it seems this is only simple replication with the master being a single point of failure. F.E.
1. update a row in the master
2. master replicates the update to multiple slaves
3. clients perform select operations against the slaves (nice load balancing opportunity)
4. the master crashes
5. No one can write until the master comes back online.
Here are the steps that seem to be missing:
6. the slaves elect a new master
7. if the old master comes back up it must realize a new master is present and become a slave.
8. clients using JDBC would need some mechanism of finding out what the new master is when an update/insert/delete fails.
Cheers.
Sun announced full Open[GL,AL] support here:
http://games-core.dev.java.net/
Here's a great example of using OpenGL/OpenAL under Win32/Linux written in Java.
(It uses the LWJGL - which is an OpenGL/OpenAL Java wrapper that uses nio).
http://www.puppygames.net/
NOTE: I'm the author of ScheduleWorld.
t ml
Some of ScheduleWorld's features:
1. Both the client and server are free
2. The client is in Java and works on Linux, Win32, and the Mac.
3. It is very secure. An interesting and short
description may be found here:
http://www.scheduleworld.com/itsYourLife.h
4. It uses Java Web Start to keep it up to date. Once you install it you will never have to install or update it again (happens automatically).
5. It uses the same format as Outlook/Lotus Notes and you can export your data at any time if you want to move to another product.
6. Outlook interop isn't quite ready yet, and the SOAP API isn't quite ready for public exposure yet but both will get there.
7. Corporate time sheet extensions (with full SR&ED - Scientific Research and Experimental Development - tax credit reporting) will be completed and ready for use in less than 2 weeks. This would be a "for pay" feature.
http://www.ScheduleWorld.com/
Cheers.
A couple of quick points:
/dev/dsp open()s to work on cards that don't provide hardware mixing. This means you can play Java sounds and run KDE/Gnome sounds at the same time.
1. FUTEX - a fast user-mode mutex. In the likely (hopefully) usual case no context switch happens when you get a futex. This can be a big win under certain synchronization loads. Imagine grabbing a futex in 15 clks vs 1,500 - 6,000 clks (these numbers may be slightly off and will vary quite a bit between machines, glibc and linux kernel versions)
2. NPTL - the New Posix Threading Library - a great improvement in both Posix compliance, performance, latency, and in the maximum number of usable threads. Ingo (one of the authors) announced the creation and destruction of 100,000 threads in one second. This is available in RedHat 9. (I don't know if futexes are available in the RedHat 9 kernel yet).
3. Graphics - Hey, I'm a Linux/Java fan and I'd like to be able to agree with you here but no. Doing a search in Sun's Java bug list will show up a lot of bugs in Linux graphics performance. In fact, check out http://www.javagaming.org/ and browse some of the posts there about it. There are tons of benchmarks of actual games that get 1/10 of the video throughput under Linux vs Win32. Sorry. (BTW, it seems the Blackdown folks are claiming an 80% 2D speedup in their latest port. I haven't tried it yet but it may help).
4. About ALSA - it rocks! What I meant to say was that perhaps RedHat could help the ALSA project support more cards and remove the completely stunned ARTS/ESD/etc... hacks and replace them with the proper ALSA DMIX plugin. This would allow multiple
(JDK 1.4.2-beta already has ALSA and DMIX support BTW - comments welcome on how well it works).
As others have mentioned the Java source is already freely available. As someone who spends most of their time in Java I can already fix bugs if I need to - having a separate license (GPL) is meaningless to me.
It would be a much better use of time and talent to make Java work better on Linux wrt:
1. FUTEX support
2. NPTL threads
3. full screen and 2D graphics are horribly slow under Linux because for some reason Sun doesn't seem to use MIT-SHM, or their pixmap caching code is doing the wrong thing...
4. Why not even spend time helping with Sun's ALSA port for Java 1.4.2.
Heck, I'm sure the Blackdown team would have dozens of ideas on how to improve the existing code base.
Rewriting from scratch? Is working together so hard? It would be such a shame to have great Linux coders work to build something that didn't work perfectly and was never used.
They do seem off. Even a 66MHz FSB wouldn't account for it. Here's an interesting post by Linus that mentions the old "int 0x80/iret" was 1761 and the new sysenter/sysexit is 641 (on a P4): http://www.ussg.iu.edu/hypermail/linux/kernel/0212 .2/0261.html
If you use the assembly language instruction RDTSC right before and after getpid() to determine how many clock cycles a simple (maybe the simplest) context switch can take you will find some interesting results:
(RDTSC gives a correct result +- 20clks)
400MHz PII: 44,000
PIV: 18,000
266MHz Athlon 1.3GHz: 6000
These numbers are from memory from my tests several years ago. I have seem posts on lkml that glibc/Linux may have the PIV case down to about 9,000 ckls but I haven't tested it. However, I am fairly certain the PII case is correct - and the 66MHz FSB was mostly to blame for the bad numbers.
Let's take Apache using on average 10 (guess) system calls for 10 x 44,000 = 440,000 clks for a 6-line index.html file.
Apache would only be able to serve 909 index.html files/second using 100% of the CPU. (simplistic)
Using Tux, we would be able to serve 10,000 index.html files/second using 100% of the CPU.
In the real world, Tux is able to server 10,000/s on my 400MHz PII, and Apache is able to server over 2,000 so perhaps it doesn't use 10 System calls/page (I was guessing). In any case, the performance difference can obviously be worth while.
Generics is auto-casting mechanism and compile-time type checking - nothing else. If you feel otherwise let us all know why.
Note: I did not say Generics was bad, or that it wasn't an improvement, or that it affected performance.
I just want to be clear: I mentioned fastutil because I feel in many cases developers would prefer to use it over Generics for these reasons:
1. It allows you to use native data types in collections.
2. It provides more algorithms
3. When used with native data types, it is enormously faster, much more memory efficient, and saves the garbage collector from potentially large amounts of needless work.
As a pro Java developer, I want to use the native 'int' type in order to save memory, have less garbage collection, and perform better. Catching errors at compile time is helpful too. I think it is unreasonable for Sun not to include specializations for native data types. If I want to have an ArrayList of 10,000 ints I should be able to use 'int'.
The link on this page states up to 10x performance but I've seen it work up to 30x performance - and you can run the code below to see this for yourself.
NOTE:
30 = 7272727/236966 where:
1. 7272727 = 2nd iteration of Int2IntHashMap!
2. 236966 = 15th iteration of HashMap (Hot spot had 12 more iterations to optimize)
http://fastutil.dsi.unimi.it/
package com.wss.utils.test;
import java.util.*;
import it.unimi.dsi.fastUtil.*;
public class TestFastUtil {
public static void main(String[] args) throws Exception {
int count = 400000;
int timerCount = 20;
long start, end;
Integer tmp;
HashMap hashMap = new HashMap(count);
for (int timer=0; timer timerCount; ++timer) {
start = System.currentTimeMillis();
for (int i=0; i count; ++i) {
hashMap.put(new Integer(i), new Integer(i));
}
end = System.currentTimeMillis();
System.out.println("HashMap put(Integer, Integer) count:" + count +
", put/s:" + (count / ((float)(end-start) / (float)1000)));
start = System.currentTimeMillis();
for (int i=0; i count; ++i) {
Integer in = new Integer(i);
tmp = (Integer)hashMap.get(in);
if (!tmp.equals(in))
throw new Exception("failed equals()");
}
end = System.currentTimeMillis();
System.out.println("HashMap get(Integer) count:" + count +
", get/s:" + (count / ((float)(end-start) / (float)1000)));
}
timerCount = 100;
Int2IntHashMap int2IntHM = new Int2IntHashMap(count);
int j;
for (int timer=0; timer timerCount; ++timer) {
start = System.currentTimeMillis();
for (int i=0; i count; ++i) {
int2IntHM.put(i, i);
}
end = System.currentTimeMillis();
System.out.println("Int2Int put(Integer, Integer) count:" + count +
", put/s:" + (count / ((float)(end-start) / (float)1000)));
start = System.currentTimeMillis();
for (int i=0; i count; ++i) {
j = int2IntHM.get(i);
if (i != j)
throw new Exception("Int2Int failed equals()");
}
end = System.currentTimeMillis();
System.out.println("Int2Int get(Integer) count:" + count +
", get/s:" + (count / ((float)(end-start) / (float)1000)));
}
}
}
It would seem to me that it's clear this patch is much more than just a non-executable stack patch. I think an important point is that while it isn't perfect, if this becomes a standard part of Linux then we can stand on the shoulders of this and tackle the next set of problems. Eventually we may get there.
If we wait until a perfect patch presents itself we will never get there. F.E. using this patch will bring to light all applications that require gcc trampolines. With time, perhaps all of the trampoline code will be replaced - several articles have been written on how to live without trampolines, perhaps this would be the push needed to change.
Nice try queuetue... BTW, it's interesting that you couldn't find a _single_ thing wrong with it. :-)