I used HFS as the HTTP server application and Firefox as the client application for a test in a production network and found a symtom which may explain "slow" reports in the forum for past years.
Test case 1: 2 laptop connected with a switch, all running 100Mbps.
Test case 2: 2 laptop connected with 2 different switch running at 100Mbps; the two switches are connected with a VPLS link with bandwidth 20Mbps.
Result 1: Single thread download speed is 30Mbps
Result 2: Single thread download speed is 14Mbps
The question: What is the bottleneck of the communication, particularly for the 100Mbps line?
During the test, I captured the packets sending out / received from the laptop running HFS server and found that in any moment, HFS will send 6 packets out from the server, and wait for 1 acknowledgment packet from the client. As a result, regardless of the speed of the link, the RTT between server and client affect the overall throughput significantly.
The follow-up question: Can the acknowledgment mechanism to be modified, say, if the previous acknowledgement is as expected, increase the number of packets send out before requesting another acknowledgment? (This is definitely not an optimal mechanism, but easy to understand)
e.g.1
a. Server send 6 pkts (1-6)
b. Server wait ack pkt (7)
c. Client send ack pkt (7)
d. Server receive ack pkt (7)
e. Server send 7 pkts (7-13)
...
e.g.2
...
a. Server send 10 pkts (31-40)
b. Server wait ack pkt (41)
c. Client send ack pkt (33)
d. Server receive ack pkt (33)
e. Server send 9 pkts (33-41)
...
