Originally by: Templer Relleg

---

So where is this "juicy" performance boost you rabbled on about?

---

Originally by: CCP Explorer

---

It's impossible to come up with a single number on the performance increase since it depends on which system we view but we are seeing improvements across the board. For example, the startup time of the server has decreased by 10% and memory usage after startup has also decreased by 10%. The new PGO Stackless Python 2.5 performs 15% better on benchmark suites. CPU load on proxy servers is 25-50% less. We had fleet battles on the test servers with about 45% more people but still comparable load on the server and players reported less lag and better frame rates.

---

Originally by: CCP Wrangler

---

You're not supposed to feel like you're logging in to a happy, happy, fluffy, fluffy lala land filled with fun and adventures, thats what hello kitty online is for.

---

Originally by: Anubis Xian

---

Not to alarm anyone, but maybe the virtual two boxing problem was intentional?

There are no legitimate ways to play the game that requires two clients being run at the same time by one person after all.

---

Originally by: Masheine

---

"Stackless" is actually oddly named, since it's not actually lacking a stack. It just changed where it keeps the stack, instead of being so heavily integrated with the hosting C program.

---

Originally by: Templer Relleg

---

Originally by: WhitePhantom

---

Originally by: Templer Relleg

---

Originally by: Niccolado Starwalker

---

I would also like to know where you got the 50% increase of CPU from. CCP have worked hard on reducing CPU. Not to increase it!

---

Im just stating a fact.
I have no idea what CCP have done, or if its my system. Yet i blame python. It could be done so much more efficient in C++ code(Or any language with a JIT compiler)

---

You really think Java would provide better performance

Like others have said they already used Python as the serverside backend, it was just updated to a newer version, the problems that were caused by it were only because of the short testing cycle and are minor at best.

---

Im sorry. I ment without a JIT compiler. C++ DOESNT have a JIT compiler.

---

Originally by: Kagura Nikon

---

Nothing on the Cpp standard prevent the implementation of a just in time compiler. And when you are coding for aserver, you know exactly the final platform. so JIT brings no gains comapred to traditional compiling.

---

Originally by: Kagura Nikon

---

I used python for years at work, including game development. And I still say, its a horrible language. Lua is far superior to game scripting. Python performance is horrible, and i find very very hard to read a code where the types are implicit defined (not explicit as in C++).

Also prefer languages were errors are detected in compile time not run time. Every huge system I worked with python we had new errors appearing months after the code was written. its good for prototyping, but would never use python by my own will on any large scale system.

---

Originally by: Kismo

---

Once I deleted the MachoNet folder from my cache, I stopped having CPU + crashing problems. I'd saw about a 20% increase in performance... maybe a bit more or less depending.

Oh, and Python does *not* suck. In fact, I've found in my time that programmers who "can't read Python" or find its usage unintuitive are pretty bad programmers. Usually the kind that are a danger to any code base.

Kismo

---

Originally by: Kismo

---

Once I deleted the MachoNet folder from my cache, I stopped having CPU + crashing problems. I'd saw about a 20% increase in performance... maybe a bit more or less depending.

Oh, and Python does *not* suck. In fact, I've found in my time that programmers who "can't read Python" or find its usage unintuitive are pretty bad programmers. Usually the kind that are a danger to any code base.

---

Originally by: Kagura Nikon

---

If you read a python function you do not know which type its supposed to receive as parameter. That means its interface is completely obscured. Clear and explicit interfaces are 99.9999% of having a readable code. Python is simply plain horrible to read. You cannot know form instant 1 what are the types involved.

Its by far the most unreadable language I ever used. In an OO language identifying the types is far far more important than reading the commands, and python fails completely on that.

---

Originally by: Erichk Knaar

---

Just a note about Python performance. Where I work, we use it for a custom load testing rig. We did this because of a few things. We can rapidly adapt this rig to almost any new server set we have to test. Like, one afternoon rapid. It performs very, very well, because we've used good algorithms and appropriate data structures. Python/Twisted allows us to use this on a potentially massive cluster (our biggest use to date was driving about 30 load generating machines) with very little work. Most importantly, it takes me all of about a day to give someone OJT on how to write scripts and work with the system because the code is so readable and obvious.

We tried a few other languages before we settled on Python. C/C++ or Java are too slow in terms of development time, for a really small performance gain (i.e. not worth the extra effort). Stuff like Perl is simply unmaintainable in a system as large as the one we have.

I would imagine that trying to re-write the Eve servers or portions of the Eve client would also suffer greatly from diminishing returns. The small increases in performance you gain will not be worth the engineering effort you expend.

---

Originally by: Kagura Nikon

---

Originally by: Erichk Knaar

---

Just a note about Python performance. Where I work, we use it for a custom load testing rig. We did this because of a few things. We can rapidly adapt this rig to almost any new server set we have to test. Like, one afternoon rapid. It performs very, very well, because we've used good algorithms and appropriate data structures. Python/Twisted allows us to use this on a potentially massive cluster (our biggest use to date was driving about 30 load generating machines) with very little work. Most importantly, it takes me all of about a day to give someone OJT on how to write scripts and work with the system because the code is so readable and obvious.

We tried a few other languages before we settled on Python. C/C++ or Java are too slow in terms of development time, for a really small performance gain (i.e. not worth the extra effort). Stuff like Perl is simply unmaintainable in a system as large as the one we have.

I would imagine that trying to re-write the Eve servers or portions of the Eve client would also suffer greatly from diminishing returns. The small increases in performance you gain will not be worth the engineering effort you expend.

---

What type of system you implement? On all systems I worked on, after I made people drop python and rewrite in C++ we got 5 to 12 times performance increase (of course there systems were bottleneck wont be related to language). And in an specific MMO client we got 7 times performance improvment and reduced memory usage by 43%.

---

Originally by: Sertan Deras

---

I think some of you have gone way off on a tangent here. Rarely is Python used to write entire systems, though it's been done. In most cases Python is used as a supplement to a system, a glue language to tie together the native components in a rapid development, scripted environment. If done correctly, Python can be a great language for this. The implicit type system requires your programmers to comment well.

If some of you are advocating writing your entire suite of game code in a native language, I hate to break this to you, but you don't know much about how games are designed and developed. You absolutley must have a system in place that allows non-programming designers to quickly change and prototype game code, without needing to get a developer involved for a recompile of the entire engine. Later down the pipe, if it becomes apparent during your profiling that a certain section of the script glue code is a performance bottleneck, then you see about adding that section of code to your native engine for execution speed. Developers capable of writing working game engines cost good money and their time is valuable. The more you can have them writing engine code, and the less you need them to write game specific code, the better return you will get on the investment you are making for that persons talent.

---

Originally by: Kagura Nikon

---

What type of system you implement? On all systems I worked on, after I made people drop python and rewrite in C++ we got 5 to 12 times performance increase (of course there systems were bottleneck wont be related to language). And in an specific MMO client we got 7 times performance improvment and reduced memory usage by 43%.

---

Originally by: Sertan Deras

---

I think some of you have gone way off on a tangent here. Rarely is Python used to write entire systems, though it's been done. In most cases Python is used as a supplement to a system, a glue language to tie together the native components in a rapid development, scripted environment. If done correctly, Python can be a great language for this. The implicit type system requires your programmers to comment well.

If some of you are advocating writing your entire suite of game code in a native language, I hate to break this to you, but you don't know much about how games are designed and developed. You absolutley must have a system in place that allows non-programming designers to quickly change and prototype game code, without needing to get a developer involved for a recompile of the entire engine. Later down the pipe, if it becomes apparent during your profiling that a certain section of the script glue code is a performance bottleneck, then you see about adding that section of code to your native engine for execution speed. Developers capable of writing working game engines cost good money and their time is valuable. The more you can have them writing engine code, and the less you need them to write game specific code, the better return you will get on the investment you are making for that persons talent.

---

Originally by: Kagura Nikon

---

Originally by: Sertan Deras

---

I think some of you have gone way off on a tangent here. Rarely is Python used to write entire systems, though it's been done. In most cases Python is used as a supplement to a system, a glue language to tie together the native components in a rapid development, scripted environment. If done correctly, Python can be a great language for this. The implicit type system requires your programmers to comment well.

If some of you are advocating writing your entire suite of game code in a native language, I hate to break this to you, but you don't know much about how games are designed and developed. You absolutley must have a system in place that allows non-programming designers to quickly change and prototype game code, without needing to get a developer involved for a recompile of the entire engine. Later down the pipe, if it becomes apparent during your profiling that a certain section of the script glue code is a performance bottleneck, then you see about adding that section of code to your native engine for execution speed. Developers capable of writing working game engines cost good money and their time is valuable. The more you can have them writing engine code, and the less you need them to write game specific code, the better return you will get on the investment you are making for that persons talent.

---

well I am game programmer. I know all this. And my experience up today is. Not worth let non real programmers do code, always at end we needed a professional programmer to review or correct a disaster that was created. So for example in current project everyone involved is a programmer or artist. The game designers are required C++ programming skills. Its working VERY nicely, much more fluid development.

---