[Standards] Proposed XMPP Extension: Stream Compression with Efficient XML Interchange

Fabio Forno fabio.forno at gmail.com
Tue Feb 19 10:21:54 UTC 2008


On Feb 19, 2008 10:02 AM,  <juha.p.hartikainen at nokia.com> wrote:
> Hi!
>
> Is there any calculations available for exi and zlib?

with exi is too difficult to setup a test environment, here is simple
dump of a session using zlib

(210, 148)
(401, 215)
(763, 375)
(835, 400)
(1063, 576)
(1469, 871)
(1585, 944)
(1795, 956)
(2123, 993)
(2240, 1057)
(2399, 1106)
(2510, 1132)
(2630, 1155)
(2720, 1192)
(4168, 1584)
(5616, 1879)
(7064, 2212)
(8512, 2493)
(9960, 2822)
(11408, 3123)
(12856, 3411)
(15636, 3957)
(15824, 4090)
(15972, 4145)
(16412, 4317)
(16561, 4382)
(17608, 4633)
(19056, 4997)
(21442, 5419)
(22479, 5474)
(23927, 6431)
(26823, 8667)
(29719, 10907)
(32615, 13157)
(34063, 14174)
(35511, 14434)
(36959, 14734)
(37992, 14787)
(39302, 15283)
(39867, 15428)
(40154, 15538)
(40457, 15661)
(40704, 15717)
(40907, 15827)
(41179, 15876)
(42627, 16232)
(43633, 16510)
(43960, 16598)
(44168, 16697)
(44295, 16730)
(44496, 16747)
(44798, 16812)
(45002, 16840)
(45072, 16859)
(46333, 16990)
(46496, 17073)
(46699, 17100)
(46767, 17117)
(47051, 17219)
(48499, 17542)
(48851, 17621)
(49053, 17653)
(49267, 17689)
(49469, 17701)
(49666, 17818)
(49868, 17827)
(50040, 17871)
(50227, 17892)
(50399, 17907)
(50597, 17942)
(50815, 17999)
(50816, 18006)
(50978, 18022)
(51192, 18034)
(51364, 18054)
(51549, 18075)
(51721, 18090)
(51951, 18161)
(52160, 18205)
(52332, 18222)
(52494, 18241)
(52698, 18287)
(52870, 18304)
(53075, 18355)
(53308, 18418)
(53470, 18437)
(53647, 18459)
(53819, 18479)
(54020, 18509)
(54192, 18526)
(54406, 18536)
(54612, 18576)
(54774, 18595)
(54933, 18620)
(55095, 18635)

On the left column the uncompressed traffic, on the right the
compressed equivalent. Comments:
- the overall compression seems low (1/3)
- but until ~ 48K it was just the login sequence and the roster &
presence burst (my roster is large... not as Peter's, but large)
- after that I had a chat with a buddy and the compression of messages
is ~ 1/7
- therefore I think it's worth optimizing the start sequence, 50K for
just logging in are a nonsense, and than with zlib we live pretty
well...

If you want more data the dump is still going on... just ping me, i
can provide also the real traffic (no sensitive data to be hidden, I'm
not planning a terrorist attack :P)

PS while I was writing this message the conversation went on, the
figures in this moment:

(82648, 23106)

(still behaving very well after the startup)

-- 
Fabio Forno, Ph.D.
Bluendo srl http://www.bluendo.com
jabber id: ff at jabber.bluendo.com



More information about the Standards mailing list