Note that there are some explanatory texts on larger screens.

plurals
  1. USJohn Robertson
    primarykey
    data
    text
    plurals
    1. COI thought StopWatch only used the usual times windows offers, e.g. with a resolution of about 10-15ms (which I would guess is really updated with each thread switch). Am I wrong about StopWatch being only a fairly course timer?
      singulars
    2. COInteresting. Where did you find that out? I didn't know that. It isn't sending that 730,000 bytes. If it sent even a fraction of that there would be no problem. When it deadlocks, it just got an ACK, sent a single push packet, and refuses to send more until it gets another ACK. If it would send just one more packet of 1460 bytes the deadlock wouldn't be there (since MS TCP would send a no-data ACK when it got the second packet). As we have played with it the size of the outgoing message (meaning how much data is posted in each socket.send command) appears important in the MS heuristics.
      singulars
    3. COActually we used wireshark to get most of the information above. We have have a very large receive window (window scaling is automatically invoked by windows since our receive buffer is large). We need speeds on the order of at least 450 Mbs over 1000Mbs ethernet. Should be doable. The receiving side primarily moves the data around in memory right now. The pauses exactly correspond to the time socket.recv takes to return, and not to the return time of any other function. It only happens in just the right circumstance.
      singulars
 

Querying!

 
Guidance

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload