Author |
Topic: Failed Simul test (Read 1066 times) |
|
chessandgo
Forum Guru
Arimaa player #1889
Gender: 
Posts: 1244
|
 |
Failed Simul test
« on: Oct 24th, 2010, 8:25am » |
 Quote  Modify
|
Hi,
so I just tried to play 6 games with CC bots to test things in view of next week's simul, but 5 of them timed out on move 2 (and the games don't appear in the database).
Was it a bad idea to start 6 bots at the same time? Server overload? I'd like to check how many games I can play in the festival simul in case there are many participants.
Jean
|
|
 IP Logged |
|
|
|
rbarreira
Forum Guru
Arimaa player #1621
Gender:
Posts: 605
|
|
Re: Failed Simul test
« Reply #1 on: Oct 24th, 2010, 8:35am » |
Quote Modify
|
I would recommend using P1 bots for this test, since CC bots cause more CPU and RAM load on the server.
|
|
IP Logged |
|
|
|
Nombril
Forum Guru
Arimaa player #4509
Gender:
Posts: 292
|
|
Re: Failed Simul test
« Reply #2 on: Oct 24th, 2010, 10:47am » |
Quote Modify
|
I have tried 4 bots (1 cc and 3 others, usually P1/P2) and occasionally had one time out. But I was playing the bots that aren't on the regular ladder, so they might have some stability issues in the first place.
I did start all 4 bots at the same time.
|
|
IP Logged |
|
|
|
omar
Forum Guru
Arimaa player #2
Gender:
Posts: 1003
|
|
Re: Failed Simul test
« Reply #3 on: Oct 24th, 2010, 11:13am » |
Quote Modify
|
Yes, it's not good to start that many CC bots at once 
It might be better to test this where you are playing both sides. Just create an open game and then join your own game. Such games are not saved, but you can use it to make sure things work.
|
|
IP Logged |
|
|
|
chessandgo
Forum Guru
Arimaa player #1889
Gender:
Posts: 1244
|
|
Re: Failed Simul test
« Reply #4 on: Oct 24th, 2010, 12:47pm » |
Quote Modify
|
Oh ok. Well sorry, I hope I haven't messed up with other things 
I was rather curious as to how many serious games I could play at once, so P1 bots or play vs myslef wouldn't work for this. Well, we'll see next week
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: Failed Simul test
« Reply #5 on: Oct 24th, 2010, 9:34pm » |
Quote Modify
|
I recall from the old days (about 2006) that I was able to play ten bots at once in some combination of CC and P2 versions. It is disappointing to hear that only six simultaneous bots crashed the 2010 server. I don't know whether this is due to something new about the server configuration, or due to the way some modern bot(s) grab resources, but it doesn't bode well for further expansion of the Arimaa server. I was expecting the server upgrade to at least double the capacity of the game room, but it obviously hasn't doubled the number of bots that can be running at the same time.
We next need to be on the lookout for whether anything besides running bots stresses out the server. I still hold out hope that many simultaneous human vs. human games will not be a huge load. Also I hope that Teamspeak and the chat room and the web interface don't suck up a lot of resources. As long as that is the case, we can always at least keep the game room up and functional by limiting access to bots.
In the past when I suggested rationing access to bots so that all the other functions of the game room (most especially HvH play) could run unabated, I was shouted down by the "everything I want must be free" crowd, and assured by optimists that more hardware would solve the problem. Now that more hardware hasn't solved the problem, it is time to think about intelligently allocating limited resources.
As things stand now, we are going to see more and more instances of all the bots being disabled to make sure that some event goes smoothly, for example a Challenge Match or the Arimaa Festival Blitz Tournament. This is a reasonable stop-gap measure, but it seems a shame that a newcomer who has wandered into the game room during a prime-time event won't even be able to play the bottom rungs of the bot ladder. Is ArimaaScoreP1 the problem here? Would it be possible to leave the lightweight bots available to newcomers while disabling the resource-hogging CC bots?
I'm just floating this trial balloon once again, but perhaps it will again be shot down by folks who are so opposed to rationing that they would rather let everyone consume every resource regardless of the outages such a policy causes.
|
|
IP Logged |
|
|
|
Nombril
Forum Guru
Arimaa player #4509
Gender:
Posts: 292
|
|
Re: Failed Simul test
« Reply #6 on: Oct 25th, 2010, 7:40am » |
Quote Modify
|
Isn't there already a mechanism that limits the duplicate copies of the same bot? (max 1, max 10, etc. depending on the bot)
I'm not well versed on the resource issues, but it seems that it would be useful to extrapolate this type of rule to the group of bots that use the most memory.
I have the perception that CC bots use more resources, but I'm not sure why that is. Wouldn't a Blitz bot use the same amount, just every other 15seconds rather than every other 2 min? And the P1/P2 bots use less only because the % time that a human is thinking is a larger ratio since the bot moves after the search depth is exhausted rather than because time is exhausted?
Anyway, I'm not against having a limit on the number of bots that are playing, especially if it prevents server and/or bot crashes. And I think it would be safer/simpler to have the server automatically protected from overload rather than having Omar available to manually make the change.
One request - if such a limit were implemented, is there a way to implement a queue? i.e., still be able to open and join the game window, but get a message saying the bot will join after x games have finished?
|
|
IP Logged |
|
|
|
rbarreira
Forum Guru
Arimaa player #1621
Gender:
Posts: 605
|
|
Re: Failed Simul test
« Reply #7 on: Oct 25th, 2010, 8:17am » |
Quote Modify
|
From this thread, I got the impression that the bottleneck on the server is mostly the RAM.
Bots seem to vary widely in how much RAM they use. I've noticed that some of the older CC bots use very little RAM, maybe because they don't use any transposition tables (which, for an alpha-beta-based bot, usually account for most of the memory consumption).
If omar wants to implement Fritzlein's suggestion on restricting which bots can be used during high-load periods, maybe it would pay off to see what bots use little RAM.
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: Failed Simul test
« Reply #8 on: Oct 25th, 2010, 1:07pm » |
Quote Modify
|
on Oct 25th, 2010, 8:17am, rbarreira wrote:| From [url=http://arimaa.com/arimaa/forum/cgi/YaBB.cgi?board=siteIssues;action=display;num=1280490150]If omar wants to implement Fritzlein's suggestion on restricting which bots can be used during high-load periods, maybe it would pay off to see what bots use little RAM. |
|
Yes, any kind of restrictions are going to be most efficient if we know which resource is the constraint and which bots use the most of it. I thought previously CPU was the constraint, but now that we have more CPU and still are constrained, it seems more likely you are right that the RAM is the problem. Understanding RAM usage of the various bots could allow for an effective selective blackout rather than a blanket ban.
|
|
IP Logged |
|
|
|
|
Home
Help
Search
Members
Login
Register
Reply
Notify of replies
Send Topic
Print