Spiral-spy uses game resources in a creative manner unrelated to the game. AFAIK, there's nothing you can do in spiral-spy that will affect the game itself.

The stub launcher, when used as (assumed) intended by Eltia, is not much different from the .bat files discussed on the first page of this thread. However, both methods allow the possibility to operate the game in unintended, malicious ways. It's hard to tiptoe around the major security risk of OK-ing the use of unofficial launchers. Very Bad Things can happen if they are abused. Unfortunately, they can facilitate larger offenses of the ToS even if they aren't considered violations themselves.

A couple notes here for those poking around:

Depending on your computer Java is already running in server complier mode by default. Any 64-bit OS or a 32-bit OS with at least 2 CPU cores and 2GB of RAM. This is called Java Ergonomics. This also implies -XX:+UseParallelGC, Xms of 1/64th physical memory and Xmx of 1/4th physical memory. All of these settings can be manually overridden. In the case of Spiral Knights the Xmx parameter is forced.

A 256mb heap size is enough for almost all of the game if running minimum display settings (including compatibility mode), but ramp up the settings and hit a boss lair and you run into trouble. An aggressive garbage collector kicks in, pausing all other threads to clear up some memory. In some areas the game wants/needs more memory beyond the set cap and you can actually watch memory usage oscillate around the 512MB mark in the Task Manager.

Increasing the Xmx value is, by far, the single most effective way around this. This allows the VM greater elbow room and the parrallel GC to quietly work in an effective manner. On a server class machines I've observed it as more than effective enough, providing enough memory was allocated in the first place. Xmx384m seems good enough for most of the game play, though some further tests are pending. Xmx512m is moving into the more than enough territory, Xmx1024m is overkill and Xmx2048m is hilarious. The trouble with aggressively increasing the Xmx value is that you can blow minimum spec users out of the water with it (though watching javaw.exe eat through more memory than a modded Oblivion game was kinda funny).

Remember, specifying any values in the Steam launcher options or web version shortcut is useless since those options are sent (if at all) to the getdown VM, which launches a new VM for the main game. Firing up the game from the command of launcher.log works a treat though, but doesn't allow Steam login.

Java/SK does run multicore (SK usage on a Q9550), however it's mostly one super thread with smaller helpers that don't necessarily make a blip on usage statistics. The way Java Ergonomics is defining server class machines implies the parallel GC ideally runs on a separate core.

According to these documents, Java Ergonomics will use the client compiler on 32-bit Windows regardless of hardware specifications. This includes running the 32-bit JVM on 64-bit Windows. The 64-bit JVM on 64-bit Windows will default to the server compiler as expected.

Kirnan, I dont know how much more I can really go, theres enough hints and tips to get you to the source, if you search the correct things on google you will find it. next google how to compile c source code cygwin and all that it requires is a dll file and renaming one file. I hope im not getting myself in trouble for saying that much.

Couldn't find anything related to the little code posted, so I investigated a bit more how steam starts SK.

With System Explorer I could see the exact parameters:
javaw.exe -Dcheck_unpacked=true -jar getdown-pro.jar . client abc Stammdateiname="D:\Programme\Steam\steam.exe"

The last part is german (Stammdateiname="D:\Programme\Steam\steam.exe") and means the parent process filepath. This is automatically added when starting a new process. I don't know how to imitate starting javaw as steam.exe, since just running this line doesn't work, since the system automatically adds another (Stammdateiname=) part at the end of the parameters.

Edit: With "doesn't work" I mean it starts not as steam client. So no steam login.

So the solution would be maybe to set this paramter by ourself.

What I noticed was that javaw.exe got called twice. So the call to the actual SK instance that runs your game, isn't called by the Users. As a result, whatever parameters you pass to the first call, may or may not get passed to the actual SK instance. (Thus explaining why we got varying results earlier before the stub was introduced, because some specified parameters never reached the actual SK instance.)

If you follow what I said above, and you have some proficiency in programming, it's not difficult to develop your own stub.

But me. But I'm having here problems compiling a complete jvm. Trying to compile openJDK my own java.exe, but I'm still at configuring the project. I think that's not the way you did it ...

So how to compile just a new java.exe/javaw.exe? I think I have the sources, but getting thousands of link fails, cause the rest of the jvm isn't build (openJDK). Also tried the jdk7 from oracle, there were also sources in it, but couldn't find the jvm.h there, that was the reason to try it with openJDK.

I would appreciate little help on this. Something I would discuss by PM but there is no such feature here.

Hi guys !
any way to increase performance on a mac running on osx Leopard ?
I constantly get stutter and lag on it, even running latest java...
I have 2 macs...one that I upgraded to Lion, and all problems disappeared ( i used to have exact same problem) and another imac that i cant upgrade to lion (because is a coreduo only )

I ve sent mail to support, but got nothing that helped...

cheers,

Thanks for the info, wish this was addressed before the steps that have been taken were taken how ever im glad to know that we can create a extra.txt file and add the parameters. I my self will be testing it out, I've been loving the lag free game play with the stub now this is a good fix I can tell people who complain to come read and try if their comfortable. Thanks again.

I have made three performance profiles (medium, very high and maximum). You can find them here. Feel free to comment on the profiles and I will make adjustment as needed.

Also if you are a Mac user and you have a profile you like to share, please post in this thread and I'll include it in the link above.

I thought Opengl is only used on Mac's version.

Again, you are free to comment on the profiles. The idea is that we create a good sets of profile and make public to everyone, and save people's time.

I feel these profiles should be provided by OOO. But at least they are listening and slowly untying our hands. Who knows maybe they are ahead of us and will provide a list of recommended profiles for people to use.

Good news is that, I tested my extra.txt on Shadow Lair. JVM used about 850MB of RAM (private + shared data) and it only lagged at the beginning of 2nd level of Ultra Snabby.

I also tried to replace the d3d options with the option "-Dsun.java2d.opengl=true", I do not notice any significant difference in performance. Looking at GPU loads, it's hover at around 12-13%. (Resolution was 1080p w/ max graphic setting.) I suspect none of the graphic acceleration option works.

I think the easiest way is for OOO to tell us what graphic libraries they use in SK and we just tweak that library's options.

I'm usually playing the game with only the -Xmx1g variable added in my extra.txt and detail settings maximized at a windowed resolution of 1360x768 on a 1680x1050 screen. I'm on Win7 x64 with 64-bit JRE7u1, so server mode is enabled by default. I'm using a character idling just underneath the fire in the Rescue Camp instance that is nearly empty for all observations. Memory valuse may be affected by how many knights were around before I switched to a quiet instance. If you really want to know full PC details just ask, but they are being held constant for each test state anyway. If I don't state an -Xmx variable it's assumed 1g.

Default State:
FPS: 100
GPU load: 25%
CPU load: 26%
Working Set Memory: 394mb (default) / 564mb (-Xmx1g)

-Dsun.java2d.opengl=true
FPS: 94
GPU load: 25%
CPU load: 26%
Working Set Memory: 633mb

-Dsun.java2d.d3d=true
FPS: 100
GPU load: 25%
CPU load: 26%
Working Set Memory: 576mb

Corsa's Setup
FPS: 95
GPU load: 25%
CPU load: 26%
Working Set Memory: 362mb

Corsa's Setup + -Dsun.java2d.opengl=true
FPS: 90
GPU load: 24%
CPU load: 26%
Working Set Memory: 365mb

My initial reaction is that trying to change the render path either has no or negative results on my system. I'm still very skeptical of anything involving Direct 3D since this game seems to avoid anything proprietary to a single operating system brand. Corsa's summed tweaks also slowed down the game overall but really dropped memory usage. Though if I've got the resources to keep things in the memory I'd personally rather use it. Then again, remember that by default on my system -XX:+UseParallelGC is already enabled as a GC tuning option since it's considered server class. That could be giving a nice boost. If anyone wants me to another option looked into I can give it a whirl on my computer for consistency's sake.

Also, remember that on any system the default flags are (as taken from lancher.log):
-Xmx256M
-XX:+AggressiveOpts
-XX:SoftRefLRUPolicyMSPerMB=10
-Dsun.java2d.d3d=false

Elita linked to a list of extra.txt files, one of them was an experimental Mac one. It contains the following:
-Xms512M
-Xmx1G
-Dsun.java2d.opengl=true

Remember that if you're on a system with around 2GB of physical memory, the -Xmx1G may be a bit too high.

I can't provide you with a concret "he said this graphics renderer" as a proof. From the time I've spent in the Technical Issues forum, I constantly see messages relating to OpenGL. There's some more clear evidence involving the use of the LWJGL, but that information is not to be discussed here. I also don't see a reason for specifying d3d false otherwise or focusing on a technology only avalible to one of the three supported platforms.

I use FRAPS as my defacto program for collecting framerate data.

@Porky Pig
How s your playing experience on a Imac ? I ve send numerous mail to OOO because I have lots of stuttering on my imacs that do not have OSX Lion on it....my specs are way above minimum requirements too...they couldnt help me at all...
I have 2 imacs, the one running on Lion works fine, what s your OSX ?

@Dogrock. It's good to know that my assumption is correct. (That your claim SK only uses OpenGL is a speculation.) It would be very kind if OOO can step up and provides a clear statement on what graphical API exactly does SK use. This way we won't have to guess and waste effort on the wrong path.

@Porky-Pig. The symptoms you described do not look good. The "Red Dots" you described are graphic artifacts that, I am guessing it's hardware or driver related. My advice would be to do a more thorough check on your hardware and driver first, before attempting any of the tweaks in this thread.

Thank you for the explanation of extra.txt file. The game is now able to utilize more memory than the 512MB limit and I've experienced a smoother gameplay with very minimal pauses (almost non-existant).

My bad, it's working fine :) Thanks for the tip.
I also want to add that Corsa's configuration has greatly improved my gameplay.

I've tested alot around and also tried different Garbage Collectors (GC) as well as different other settings.

I've had big problems with the standard configuration, like every 30 minutes a 1 second freeze in arena. I died like 1000 times from this. And now after some investigation with VisualVM and some configurations I come up to some solutions.

Read this to an end, because there are some informations you should know.

I'm using the Sun JVM, and I think the most people, too, since it is the first result when searching for java. When it is stated as "better for Multicores" it better to have at least 3 Cores, with 2 Cores it doesn't really matters).

Some short rules:
- Don't use "-Xincgc" with this gc solutions or mix the gc solutions.
- Steam doesn't use you installed JVM, instead it has it's own, a really old one.
- If using G1 make sure to have Sun JVM 1.7 or above in the java_vm directory in you steam spiral knights directory.
- Don't overdo it with RAM and the disableGC option, else you could have a big surprisingly freeze.

GC Solution 1 - CMS:
This is the Concurrent Mark Sweep GC. It's not running too long, so you just have small freezes, like 100-200 milliseconds. It checks incrementally also the Tenured Generation and tries to reduce it, so it works against a Full Collection. Still when a Full Collection is done it freezes. This is recommended for single core CPUs, since Parallel can't help then.

-XX:+UseConcMarkSweepGC : Setting it as GC
-XX:+UseParNewGC : Adding a parallel GC for Young Generation Objects (better for Multicores) (specially for CMS, don't use it with other GCs)
-XX:+CMSIncrementalMode : Decreases the freeze time again, by doing the concurrent phase incremental.
-XX:+CMSIncrementalPacing : Enables automatic adjustment of the incremental mode above, based on statistics collected at runtime.

GC Solution 2 - Parallel:
This should be the standard configuration when using Sun JVM 1.5 or above, except for the UseParallelOldGC option. For each core it creates one thread for the GC, more than 1 thread per core are decreasing the performance, so you shouldn't do it. So you don't have to adjust the number of threads for this GC.
It still just freezes when doing collections, but it should do them much more faster. I tested it in FSC when I got constantly freezes with default configuration and it worked for me like a wonder, it reduced the occurence rate by like more than 90%. If you have a good multicore CPU try this.

-XX:+UseParallelGC : Setting a parallel GC for Young Generation. (better for Multicores)
-XX:+UseParallelOldGC : Setting also a parallel GC for the Tenured Generation. (better for Multicores)

GC Solution 3 - G1:
This one is still in experimental phase (like a Garbage Collector in beta phase) so use it with care. It only works when having a SUN JVM 1.7 or above, else SK just won't start with this options. It used a lot more of your CPU but does also a good job in reducing freeze time for me. I use this as my standard configuration. This GC can't be explained with Young and Tenured Generation Collections since it is a completly new method of garbage collection.
I can say it frees constantly the RAM and Full Collections occur as often as being hit from a thunderbolt.

-XX:+UnlockExperimentalVMOptions : Enabling experimental mode in JVM
-XX:+UseG1GC : Setting G1 for GC.
-XX:MaxGCPauseMillis=50 : Setting the maximum freeze time in milliseconds (still intervall has to be set)
-XX:GCPauseIntervalMillis=1000 : Setting the intervall in which the above maximum millicesonds are allowed to occur.

So in this example the G1 tries to run for 50 milliseconds in one second. Setting this to 0 doesn't mean you won't get freezes ... btw 0 is a bad number for options.

Extra Solutions:
Explicit Collections
To adjust the Explicit Garbage Collections which comes from the application (so from OOO devs) you have 2 options.

-XX:+DisableExplicitGC : This disables them completly, so nothing is done then the application commands a GC.
Only when using CMS:
-XX:+ExplicitGCInvokesConcurrent : This invokes a Concurrent GC for the CMS instead of a Full Collection.

Since I use G1 I just can turn em off, so I use the disable option. Since I don't like freezes.

Decreasing Loading Time
I have found an option which really does a good job on that.

-Xnoclassgc : This never collects classes, so they remain in Permanent Generation, which is about 20MB and doesn't really grow much bigger. No negative effects until now.

Garbage Collection explained:
All the RAM for the application is divided into the Young Generation and the Tenured Generation, and also the Permanent Generation. The Garbage Collection frees RAM from objects which is not used anymore from the application. When a GC occurs the application is completly stopped, in default configuration.

Young Generation: All new created objects are firstly stored there (in default configuration (from OOO) it's about 60 MB) and it fill very fast like, 15 MB/sec, just by standing around in town. When it is full, a GC starts to clear the Young Generation and stores the remaining objects which are still alive in the game (referenced) in one 1 of 2 special parts of the Young Generation (Survivor Space). If the Survivor Space is full it is also garbage collected and still living objects are copied into the Tenured Generation. This GCs are mostly not even mentioned if the Collections are not too big, like 100 MB is ok.

Tenured Generation: The longer living objects are stored here. It's about 170 MB in the default configuration of SK. If it is full a GC occurs and it can take much longer than in the Young Generation and causes the big freezes. I heard of 10 second freezes from some people. On my hardware it causes around 1-2 second freezes. When many new Objects are created, like entering a new level or going through the arena this Generation increases fast. If the Young Generation GC Rate increases (e.g. cause of much monsters and battle in Game) this Generation increases too fast with young living objects, which doesn't really belong here, but they didn't had enough time to die while beeing in Young Generation. Increasing this part will deacrease the rate at which the freezes occur, but the freezes will get worse, since the GC has more to RAM to clear.

So Until now we know of 3 Situations when GC is invoked.
When Young Generation is full, a GC is started for the Young Generation and stores the survivors in Survivor Space.
When Survivor Space is full, a GC is started for Survivor Space and stores the survivors of this GC in Tenured Generation. This occurs when Young Generation is full. So 2 GCs are done in a row.
When Tenured Generation is full, a GC is started for the Tenured Generation. This occurs when objects are stored from Survivor Space, so 3 GCs are done in a row. This is known as Full Collection.

But there is still another Situation when a GC occurs. That is when the applications commands it. Then an Explicit Collection is done, this is like a Full Collection but on command, so it does not check wether it is needed or not. This is done very often in SK.

Now I hope you have learned something and you know better what you are doing there with your options.
Just increasing RAM with some GC could make the freezes worse, so better test it and test it for more than just running through Haven.

Now all* my wisdom belongs to you.

*not really

Thank you for sharing these information. I have put up two more profiles for the parallel GC and G1 GC (with credit attribute to you). Feel free to check them out.

-Xnoclassgc works wonders. Map loading is now almost instant.

The Java 7's G1 also seems to work very well. The drawback seems to be that it uses more CPU cycles.

One of my guildmates who isn't into tweaking, just played the game with the new OOO tweaked setting last night. He reported the game appears to be smoother now.

I noticed my nVidia driver has stopped applying Antialiasing to SK and I couldn't get it to work anymore (even with nVidia Inspector). Is it just me or other people observing the same thing?

Copy the file to Steam SK folder before you start Steam.