The problem of occasional long response times for PID resolution could
be solved by forcing the resolution system for a prefix to use only TCP
for all necessary communication.
These occasional long response times for PID resolution are due to package
losses caused by the cross atlantic WAN communication between the Handle
Proxy Server, the Global Handle Registry, and the different Local Handle
Server responsible for resolution. TCP is in comparison to UDP in such a context a more
reliable protocol to handle these package losses but has the drawback of
a bigger communication overhead. In order to significantly lower the
ratio of long PID resolution times in the case of prefix 11858 it was
decided to accept this and guarantuee the use of TCP at all levels of
the PID resolution. This could be a test case for other prefixes as well.
As a consequence we now have much less PID resolution times above 5sec,
from before 0,025% to now 0,0008%, and less PID resolution times between
1sec and 5sec, from before 0,025% to now 0,0008%. The tradeoff is, that now the minimal response time for resolution increased from before 0,32sec to now 0,55sec.