Comment 5 for bug 386199

I've attached a better trace now. It contains a full Evolution session from start to finish, and excludes X11 traffic from the rest of the desktop. I used `netstat' to see which clients were connected to the Xephyr server before starting evo, fired up a tcpdump, started evo, did the testing, used 'netstat' to see what clients were connected now, and exited evo. I then diffed the two netstat results, found the new client source ports, and filtered the dump to include only those (sport: 37625, 37627).

The compose window display took 17 seconds from clicking on the "new" button to the finish of the window drawing. The time range was 11:37:30-> 11:37:47

When filtered for evo traffic only things look a fair bit different; see the I/O graph I'll attach shortly. There are several distinct phases apparent in window creation, each of which take several seconds. I'm not sure what's going on yet, and it's far from clear that network round trips / throughput are even the issue.