Stress Test - 2008-05-10 at 20:00 GMT

[Cebot]

The devs plan another stress test on Laanx, because the last one was announced too late on the forums.
So before it gets forgotten again, here's the reminder.

Next Stress Test is on 2008-05-10 at 20:00 GMT - Please everyone log in with as many clients as possible. It is really important to the devs, so take it seriously please.

[neko kyouran]

moving to the general discussion area for now since more people will see it their than here.  will move back to the server area once it's completed.

[Knallgrim]

Hm, you should post this as MOTD too, if it isn't already.


Where do we meet btw? Any place in particular?

[Xillix Queen of Fools]

I am going to suggest the DR citadel because it is easy to get there from anywhere.

[Kerol]

Summarizing from during the last stress test:

[Sa Mai 3 2008] [22:17:38] <Khaki> hmm, network performance just dropped
[Sa Mai 3 2008] [22:18:28] <Xordan> everyone is jumping
[Sa Mai 3 2008] [22:18:49] <Khaki> ah cool
[Sa Mai 3 2008] [22:19:10] <Kerlap> i'll paste the ping from all i got later
[Sa Mai 3 2008] [22:19:26] <Kerlap> from what i can see it's between 360-500
[Sa Mai 3 2008] [22:19:36] <Xordan> with the odd spike
[Sa Mai 3 2008] [22:20:22] <Kerlap> Khaki, what says the server?
[Sa Mai 3 2008] [22:20:59] <Khaki> psserver 9%, npcclient 16% cpu
[Sa Mai 3 2008] [22:21:24] <Xordan> pretty good.
[Sa Mai 3 2008] [22:21:29] <Khaki> 60% 28.6% memory

[Sa Mai 3 2008] [22:25:17] <Khaki> it was sending 53 messages per client on average
[Sa Mai 3 2008] [22:25:43] <Khaki> to about 39 clients or every call to the network

Regretfully we only had 260 players online at max, which wasn't even more than the 280 we had online once during peak time.
There were some modifications of network code and attempts to fix memory leakage server side, which we'd like to see if it works well next time.
We also were testing the performance of the new server this way which was quite successful as you can see in the chat (except the memory usage).
I extracted the pings from the chat logs which you can see beneath. These are the pings the clients had to server during the test. Sorry for the format but I think it's more readable as list.

The maximum number of allowed connections is currently set to 1000, so number  of players online shouldn't really constitute a problem.
Hopefully we'll reach the 300 players online this time - and next time the 500


BTW: The 256 player limit was an April's Fool.


Pings extracted for those interested, from last time:

 187
 277
 279
 425
 285
 333
 303
 377
 456
 528
 383
 306
 422
 351
 389
 356
 402
 307
 364
 388
 269
 287
 1931
 374
 399
 549
 306
 419
 421
 345
 287
 306
 433
 323
 423
 292
 425
 344
 452
 424
 424
 516
 473
 456
 356
 263
 352
 355
 391
 399
 388
 279
 461
 340
 450
 333
 406
 412
 412
 283
 439
 364
 330
 416
 412
 399
 433
 423
 356
 333
 366
 305
 409
 285
 324
 478
 431
 415
 311
 397
 397
 380
 384
 969
 550
 311
 428
 971
 285
 411
 397
 307
 403
 557
 544
 314
 387
 361
 436
 309
 358
 339
 354
 342
 300
 455
 345
 348
 358
 468
 317
 435
 478
 442
 478
 442
 380
 285

These numbers given, Aiken calculated average ping 402   standard deviation 175
Removing the 1931 from the list gives an average ping of 390  and standard deviation 103

[Cebot]

Don't miss the stress test! it starts in about an hour
Log in with as many clients as possible!

[Nikodemus]

ooooops ;P

[Mythryndel]

I was on for the last one... quite by accident... but RL got in the way of this one...

Is there another one planned soon?

[Kerol]

RESULTS:

We expected more people. There were only ~250 players online max, 95% of them in DR, thus it was even less than the first time - even though we announced it much earlier!!!   

I got 99 messages about ping. One had a very high ping (someone other than last time), rest was medium:
<Aiken> with the 1904 ping ,  avg 384.101013184     stddev 168.636749949
<Aiken> without the 1904 ping,   avg 368.591827393  stddev 70.111549128

Quote from analyzing the test afterwards:

[22:49] <Khaki> 'Resending non-acked packet' is printed everytime >300 packets are resent in one iteration and on every 200 iterations
[22:49] <Khaki> and that log is only over 10 seconds
[22:49] <Khaki> so we're resending a lot
[22:50] <Khaki> would be nice to have a tool that could connect to laanx remotely and measure these statistics
[22:51] <Khaki> at peak time, when everyone mass logged off, we had 13,000 resent packets
[22:52] <Khaki> Sun May 11 04:36:35 2008, Resending non-acked packet statistics: 789.425 average resends, peak of 13531 resent packets
[22:52] <Khaki> Sun May 11 04:37:20 2008, Resending non-acked packet statistics: 2178.55 average resends, peak of 13531 resent packets


That was the main reason for doubled chat messages, ghost accounts etc.
CPU/mem usage was similar to last time.

We're looking into means to reduce lag on client side (graphical) and reducing the amount of lost net packets.