Any clues to source of this delay?

[Skip Montanaro]

    Markus> Problem:

    Markus> On Linux, single connections have 'glass ceiling' of 50
    Markus> roundtrips/second; no matter what I kludge, it seems to stay
    Markus> there. Only way to fix this was to send 'empty' packets after
    Markus> processed ones, thus increasing throughput by about 5X.

    Markus> On NT, the glass ceiling is same, just 1/10 of Linux's (5
    Markus> roundtrips/second).

TCP's slow start algorithm, perhaps?  If you can figure a way to reuse the
connection for multiple messages you may find that subsequent messages are
sent more quickly.  A quick check at Google suggests that slow start details 
might be found in RFC's 2001 and 2581.

Skip Montanaro	| http://www.mojam.com/
skip at mojam.com  | http://www.musi-cal.com/~skip/
847-475-3758