When are you planning to upgrade the system?

[alex[LSD]]

There were about 300 jobs in queue (I m talking bout July Monthly Contest). I had to wait the whole HOUR before having my problem checked. I consider it slow. For example, check out the acm.sgu.ru site. They got time limits 0.25-1 second on most problems. The inputs are small but it doesnt make em easy to solve.

[sohel]

The system administrators and other concerned people should think seriously over this matter.

We must know about the verdict of the problem we send ASAP so that we can make corrections if needed. For Example in the last contest I sent one problem and the verdict came after the end of the contest. Sadly the reply was WA... but after debugging I noticed that I made a minor mistake in the coding...
.. and I was unable to solve it during the contest.

[enjolras]

Hi,

We planned a system upgrade long time ago. But being a non profit project, makes difficult to find the money to do so. We have recently reached an agreement with a non profit foundation here at the University so we can receive donations from users to finance new hardware, new software, and the people behind this. Go to http://online-judge.uva.es/problemset/ and you will find it better explained clicking on the "Read Me First" link in the Donate!!! box. 50000 euros looks like a lot of money, but there are about 10000-15000 active users, so it's only 5 euros each.
The idea is to have a new system up and running next october. Probably with a newer and faster judging software, and surely with a really big server, that should allow immediate responses.

Regards.

[shahriar_manzoor]

The difference with acm.sgu.ru and UVa site is that they set time limit per test case (I think) and in UVa we set time limit per input file. So naturally our time limits are high. The norm of ACM ICPC is like time limit per input file. but all russian contests generally set time limit per test (At least it was like this when I was a contestant and I also find that sometimes more useful). Morever we cannot set time limit in fractions of a second (The judge is designed in this way).

We would prefer a powerful judge and much higher time limit (as it is done in Real contests of ACM) but limitation of budget seems to be a problem.

I try to keep time limits low but some times it becomes too tight and some time it becomes too high. But I am quite sure that there will be no such trouble in Regional warmups as we will try our best not to include any easy (or looks very easy) problem. I hope the problem setters help us in that fashion .

[alex[LSD]]

Yup! My bad, Shahriar you re right. MOST of their time limits are per-test time limits. But still instead of 100 testcases one could choose 30 but critical ones, ones that have non-zero probability to detect incorrect solutions. Still it seems to me that on SGU site they don't usually have 100 tests per problem...

[shahriar_manzoor]

There are four issues here

a) When u set time limit per test not only the time limit is low but also you can exit if only one test fails so in that sense it is more optimal.

b) When you allocate time limit per test the option of including many test cases becomes impossible. Because for small and large cases you have to allocate same time hence 100 test cases would actually take actually time limit equal to 100 large test cases which is very high.

c) For most problems it is possible to set small data but some problem requires 1000000 numbers as input just to ensure that a counting sort passes but a qsort() doesn't etc.

d) The most difficult part of problem setting is making critical data. for many problems one way to get away from this is generate many test data and thus be sure that all possible data has been generated. All these are not because problem setters are lazy but because they have less time. Setting a good problem can often take 3 to seven days (Hard to believe but true). Because when a contestant solves a problem he knows that it is solvable but when problemsetters do it they often find that such a problem is not solvable or requires too much time or the initial idea was wrong etc. In this way judge data becomes large.

I hope it explains many things...

[alex[LSD]]

I believe there still can be a compromise.

a) I told you! I like that system better.
b) Well I can't help believing that short tests will pass fast on MOST solutions. Besides, WHY does it have to be 100 test cases??? 8-O
c) Yes... but you don't have to check it 100 times
d) Can't say nothing here, no, you people aren't lazy.

P.S. Perhaps, this all here ain't my business.

[shahriar_manzoor]

Well I think it is everyones business as long as the discussion is healthy. What I wanted to say is that whatever measures we take there will always be some problems with higher time limit. But if the judge is upgraded then the time limit is becomes low, so very soon the judge needs to be upgraded. It is very awkward for me to say to the problem setters "change your problem statement and limit because we cannot give you 5 second time limit and so on...", the slow judge is hurting everyones freedod very much. For example someone (Who is a very good problemsetter) wanted to contribute problems for UVa and when I sent him the list of restrictions he never mailed me back (The reason may be different but I assume this reason).

[Tahseen Mohammad]

Shariar bhai

You are absolutely right. There will always be some problems
with higher time limits & you cannot do anything about it. But for a contest
of 8/10 problems you can set lower time limits say for 5 problems that will
imrove the response time tremendously.

And surely Upgrade of the system is necessary no matter how cleaver
solution we devise.

And uva's slow response is also caused for its immense popularity. No
other site or their contest gets anywhere near the hits/submission as uva
does. Go its bad because its so good.

So let us wait for the golden moment when we recieve instant reply to
our submission in a uva online contest.

[Per]

There's another way to reduce the judge load, namely setting harder problems. This was very evident in yesterday's contest, where there were many trivial and next to trivial problems.

[little joey]

Hmm. When I was staring at the judge's status screen yesterday, waiting for my submissions to compile and run, I got the impression that only a small fraction of the host's time is spent on the actual compiling and running. I guess the most time is spent on book keeping, producing online statistics and most of all, handling web-requests. So setting tighter time limits won't help much to increase the response time. I think it would help much more to distribute the different processes over different machines. I bet my old 386 could beat the judge yesterday in compiling and running submissions if it were the only tasks it had to do.

Of course you can de-popularise online contesting by making the problems only solvable by the world's elite (sarcasm not meant personal, Per ), but that is not the aim of a local contest. In a local contest also the 'lesser gods' should be able to participate and score a point or two, I think (and that's not meant to be patronising). Participating in a contest should be fun, and there's no greater satisfaction then seeing your submission getting Accepted.

[Per]

I guess the most time is spent on book keeping, producing online statistics and most of all, handling web-requests. So setting tighter time limits won't help much to increase the response time. I think it would help much more to distribute the different processes over different machines.

But isn't that already the case? Two machines, one judging programs, and one handling statistics, generating web-pages, etc.?

Of course you can de-popularise online contesting by making the problems only solvable by the world's elite (sarcasm not meant personal, Per ), but that is not the aim of a local contest. In a local contest also the 'lesser gods' should be able to participate and score a point or two, I think (and that's not meant to be patronising). Participating in a contest should be fun, and there's no greater satisfaction then seeing your submission getting Accepted.

Yes, of course, I did not mean that the contests should consist only of Really Hard Problems(tm). There should always be one or two simple problems. But yesterday, there were something like four or five simple problems, which generated quite a lot of submissions, and there were no hard problems (except possibly Problem I, though it was more messy than difficult )