Re: [Server-devel] more 0.5.2-dev02 ejabberd testing
2009/3/6 Daniel Drake d...@laptop.org: I will continue testing by rebooting the server and booting all the XOs again, checking presence visibility, but right now they need a charge, so that will probably have to wait until monday. OK. Booted the server again, and all XOs, and they all managed to see each other just fine. ejabberd memory usage: 0 35124 12 53252 24 64816 36 80036 48 93512 60 108244 72 127000 Also checked olpc-xos output and manually counted the icons at various points, no problems. I then launched a Record activity and shared it between all 72 laptops. It worked very well. However, after shutting a few machines down, some remaining XOs were missing about 80% of the XO icons on the neighborhood view, until sugar was restarted. olpc-xos output was correct and showed all. I think there is probably a sugar bug causing disappearance of icons after the rings of users around activities have disbanded :( Daniel ___ Server-devel mailing list Server-devel@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
[Server-devel] more 0.5.2-dev02 ejabberd testing
I ran more ejabberd/presence tests with 0.5.2-dev02, this time with 74 XOs. My process: - NANDblast all XOs - delete all ejabberd accounts from XS (left over from the last test) - turn 12 XOs on - connect all to AP, register to XS, restart sugar - confirm that ejabberdctl connected-users reports expected number of users - take one or two XOs (out of the whole 74), check olpc-xos count of neighbour XOs, and attempt to count that the number of icons on the neighborhood screen matches - record ejabberd memory usage using ps_mem_py Results: all success It became hard to count the XOs from 48 onwards, so the counting test became more of an approximation. As you might expect from complex software, the memory usage went up and down a bit, for example see how ejabberd shrunk between 72 and 74 users, or the small increase between 48 and 60 ejabberd memory usage (first column is number of connected XOs, second is memory usage reported by ps_mem_py) 0 32788 12 46828 24 61248 36 74924 48 90128 60 93860 72 12 74 116536 Then, we decided to have some fun and share a write activity between all laptops. It worked very well but not perfectly: - as the XOs started to render rings of users on the neighborhood view, and as those circles grew, the clickable document icon disappeared off the edge of the screen on some laptops - the syncing between XOs was very slow after 50 users, but it did work - after returning from lunch, the write activity had disappeared on many laptops (known bug) ejabberd memory usage with about 60 XOs in the same activity grew to 138700 Then I turned all the machines off and checked the ejabberd memory usage: 83612 Some ran out of battery power during lunch, and it seemed to take a lot of time for them to time out from ejabberd. I will continue testing by rebooting the server and booting all the XOs again, checking presence visibility, but right now they need a charge, so that will probably have to wait until monday. Daniel ___ Server-devel mailing list server-de...@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
Re: [Server-devel] more 0.5.2-dev02 ejabberd testing
On Sat, Mar 7, 2009 at 7:24 AM, Daniel Drake d...@laptop.org wrote: I ran more ejabberd/presence tests with 0.5.2-dev02, this time with 74 XOs. ... Results: all success That's excellent news. It became hard to count the XOs from 48 onwards, so the counting test became more of an approximation. As you might expect from complex software, the memory usage went up and down a bit, for example see how ejabberd shrunk between 72 and 74 users, or the small increase between 48 and 60 Yes, ejabberd's per-user mem usage is higher at initial connection time. It is still using ~1MB per user, which is disappointing. Part of that goes to the ssl libs, part to the @online@ roster. Some ran out of battery power during lunch, and it seemed to take a lot of time for them to time out from ejabberd. Good point - I'll see whether it's posible to shorten the timeout there. Outstanding report -- thanks! martin -- martin.langh...@gmail.com mar...@laptop.org -- School Server Architect - ask interesting questions - don't get distracted with shiny stuff - working code first - http://wiki.laptop.org/go/User:Martinlanghoff ___ Server-devel mailing list Server-devel@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
