what is the cluster result?

what is the cluster result?

Hi everybody,

As I sent some test on the cluster, I wonder what does it compute exactly?

My first idea was that you based on the default folder architecture and you run our programme with scenario available in scenario folder. But looking at the time result, I suppose it doesn't work like it.

Do you run our programm with other flight and alliance files?

In this case, is it possible to have this kind of files? I already see that there is another set of file available on the forum but I hoped to get a bigger one.

If I ask this question, it is because the last "make submit" I did compute the result pretty quickly and I wonder if it is due to an issue or not?

Regards,

21 posts / novo 0
Último post
Para obter mais informações sobre otimizações de compiladores, consulte Aviso sobre otimizações.

Hi,

We will add bigger files on the cluster soon.We will also give some of this files on the forum. You just need to have in mind that you will not have the files we will use to evaluate the programs at the end of the contest. We don't want contestants to adapt their programs to our test sets.

Hi,

thank you for you answer.
It is normal that you don't want to share files you will use to evaluate the programm. But, is it possible to know if the output file generate by our programm on the cluser is correct or not?

Actualy, the web page address that we gave you by email already provides this information. I can see that for you the result is "OK - final step". This means that you succeded the last benchmark available on our cluster.

I'll try to provide some new tests this weekend or next week.

It is really strange because for one test, I removed a field in the output file. So, I know that the output result was an error be the message was "OK - final step".
Thank you for providing new tests :-)

In fact, I let you read this !
The brenchmark is not testing your output. The goal is just to have an idea of what are your performances ! ;)

Regards,

Timothé Viot, Engineer student, Insa Rennes 1 France

The benchmark test the output on our examples. I'll check on the cluster to see if there is a problem or not. Could you send me the timestamp of the test that should have fail please by email ? I will check it.

edit : You are probably talking about the CompileScale2 test ? Because I can see that your compileScale3 failed.

This is normal, the compileScale2 has no output. You don't have this test file and the goal for you is not to cheat on it. It just test a program call with no solution. Take this test seriously because we could eventually test this kind of scenario.

Hi Cédric,
After the cluster is working again I got an error in the CompileUnitTest, but there is no feedback what exactly went wrong...

"CompileUnitTest ERROR - Output error 0.01 434 mn 2012-11-22 02:00:01 1353546001"

A few days ago I got a compiling error in CompileScale2 ("ERROR - binary could not be built correctly") and the output of make was provided.
Could you please tell me what went wrong this time?

Rock the bits!

You should get the message : Command terminated by signal 11

Which means that you are having a segmentation fault. This problem appears on the CompileScale2. Do your program succed the newinput2 ? This test is almost the same size that the CompileScale2.

Yes, it succeeds all provided files in the reference, and extra 2 scenarios you uploaded + the scenarios I generated.
I will upload a new version now...
My best idea to locate the segfault is commenting out code and submitting until it only gives an output error, but not a segfault and then see what could cause it.
Or could you think of a nicer way to locate the segfault?
It is probably not possible to run it with valgrind on the benchmark right?

Rock the bits!

The best way to locate segfault is to find test (on your computer) on which your program fails. Sometimes its possible to locate error just by reading code, but I suggest to write your own test generator.
Good luck!

It's not possible to run valgrind.

I don't know if it can come from that point but when I run your makefile, I get :
make: warning: Clock skew detected. Your build may be incomplete.

run a full make clean and remove all unnecessary files before submitting. It's possible that your project failed the build and you should get some new error messages.

Hi,

I have some questions to the benchmarking cluster:
-Is it possible to reset the benchmarking cluster results and start from CompileUnitTest again?
-What does CompileScale4 mean? Is this the last step?

Thanks!

You can submit code that doesn't compile and then submit valid code again to reset. (The reason they start from the last passed benchmark again is probably that it saves time, so the submissions don't take that long to run and the queue of pending submissions doesn't get too long, so just do it too often.)
Yes, Cédric said he will add a new benchmark ;)

Rock the bits!

Citação:

Hannes T. escreveu:

-What does CompileScale4 mean? Is this the last step?


No, for now the last step is CompileScale5. But more tests might be added later (next weekend I suppose, but it only depends on contest organizers).

The benchmarking cluster is now running for 178 mn on our code. Can someone please kill our program? Why does the Timeout not work?

Why am I getting so low cpu usages?
Like this one: CompileScale2 ... (CPU usage:43%)

Rock the bits!

I suppose it's because of overhead of creating/syncing threads. Try to submit solution without any parallelization, I'm sure you'll get something about 90-98%.

It is already sequential...
Is it possible that I/O operations slow it down that much?
Are the input files stored on an HDD, SSD, or already cached in RAM?

Rock the bits!

Citação:

Heye (aka. slevin7) escreveu:

It is already sequential...
Is it possible that I/O operations slow it down that much?
Are the input files stored on an HDD, SSD, or already cached in RAM?

Hello slevin7, I higly doubt that you have the input prepared in RAM. They are most probably on HDD, maybe SDD, I don't know what they're using for permanent storage. Unless the I/O operations take up most of your program running time, I don't think that's the cause of your low CPU usage. Maybe the cluster is overloaded, so you don't get your full share - I'don't know if the scheduler waits for enough completely free cores needed for your benchmark. It would be nice if someone could officially confirm this... Best regards, Nenad

Citação:

Heye (aka. slevin7) escreveu:

Why am I getting so low cpu usages?
Like this one: CompileScale2 ... (CPU usage:43%)

We have got different times and CPU usages at different times of day when we submit the same code multiple times... We think it depends on the benchmark's loading during a day.

Deixar um comentário

Faça login para adicionar um comentário. Não é membro? Inscreva-se hoje mesmo!