Post your hubstats

[Toast]

Pretty useless thread but it could be fun to see

Code: Select all

[21:48:11] <-Future-> Death Squad Hub. Version DSHub Theta.
  Running on Windows XP Version 5.1 on Architecture x86
  Java Runtime Environment 1.6.0_05 from Sun Microsystems Inc.
  Java Virtual Machine 1.0
  Available CPU's to JVM 2
  Available Memory to JVM: 66650112 Bytes, where free: 2989392 Bytes
Hub Statistics:
  Online users: 442
  Connecting users: 55
  Uptime: #4h#45m#25s#61.
  Bytes red per second: 849.5
  Bytes written per second: 70834.6

[Pietry]

Code: Select all

Death Squad Hub. Version DSHub Theta.
  Running on Windows XP Version 5.1 on Architecture x86
  Java Runtime Environment 1.6.0_01 from Sun Microsystems Inc.
  Java Virtual Machine 1.0
  Available CPU's to JVM 2
  Available Memory to JVM: 66650112 Bytes, where free: 2205912 Bytes
Hub Statistics:
  Online users: 707
  Connecting users: 103
  Uptime: #4d#20h#20m#22s#479.
  Bytes red per second: 6070.7925
  Bytes written per second: 102481.695

[Dj_Offset]

Fri, 24 Oct 2008 18:36:27 +0000 INFO: Statistics users=81, net_tx=1 KB/s, net_rx=0 KB/s

[Dj_Offset]

Thu, 30 Oct 2008 22:52:19 +0000 INFO: Statistics users=7896, net_tx=10803 KB/s, net_rx=8 KB/s

[Pietry]

That is real ? do you have some address we can check out ?

[Dj_Offset]

That is real ? do you have some address we can check out ?

Very much so.
However, it was done with a benchmark tool I am writing which emulates real users, and tests performed on two linux computers on a 100Mbit local area network.

Some tweaking were needed in order for the hub to allow for a very large send queue.
It peaked at about 250MB memory use.

About 20-25% less bandwidth and memory usage if "low_bandwidth_mode" is enabled.

[Toast]

Soft: eHub
Soft version: 0.4.0
Users: 1016

eHub passed 1000 users without lag to bad i dont have more info on CPU usage and bandwith usage

[blastbeat]

hi,
at the moment i try to test and optimize the possible usercount of luadch and wrote a little script
to connect multiple times to a hub. you can download a package for windows here:

Removed link by me. // Pietry.


i tested it with luadch, dshub kappa rc1 and uhub-0.2.3-3429 on winxp, amd 3000+, 1 gb ram via localhost. the settings in hubstress.lua were

Code: Select all

local bot_count = 2000
local bot_timeout = 0.9
local bots_per_second = 10

that means the script tried to connect 2000 bots to the hub and the rate should be 10 bots per second. here my results:

dsub started to drop and disconnect bots when about 500 bots were connected; i stopped the script
when only 180 bots remained in the hub. the cpu count were high, ram usage about 90 mb. screenshot:

http://home.arcor.de/blastbeat/hubstress/dshub.JPG

uhub had a great performance, minimal cpu and ram usage, but at 1270 bots the hub crashed:

http://home.arcor.de/blastbeat/hubstress/uhub.JPG

i tried the last luadch svn and didnt reach the 2000 logins, so i played a bit with the garbage collector but the result was only marginal better:

http://home.arcor.de/blastbeat/hubstress/luadch.JPG

at least all bots connected, but more users means more objects in luadch and this seems to crash the garbage collector

[Pietry]

DSHub has a problem with cpu usage as I tested yesterday with 1000+ real users. If the users are connecting too fast indeed then the hub hooks up ( ? ) , but if they connect slowly a higher limit may be reached at the expense of cpu usage ( 60-80 % ) which is not normal . I believe that my first parts of the code ( when i was learning java ) is quite bad and needs rewriting. And the 1000 user hub is not stable ( freeze up sooner or later, yesterday it lasted more than an hour, but it's quite unpredictable ).
If luadch and uhub have higher performances it's very good for the adc hubowners , they have good alternatives, but still there is no real usable and reliable hubsoft that given the proper resources ( ram, cpu, bandwidth ) to support a high user count stable. But this is just fine tuning that can be done, just that its a bit tricky and requires lots of testing. I;m happy to see luadch is handling better, hope you will make it even more.

[Pietry]

Would be interesting to test ADCH++ as well...