linuxdcpp-team team mailing list archive
-
linuxdcpp-team team
-
Mailing list archive
-
Message #09027
[Bug 1909853] [NEW] Don't use custom socket buffer sizes by default
Public bug reported:
Modern operating systems (Linux 2.4+, Windows Vista+) should support TCP
tuning (https://en.wikipedia.org/wiki/TCP_tuning) to enable higher per-
thread speeds with fast connections. However, the current default socket
buffer size (65536 bytes) set by the client will prevent the feature
from being utilized.
Example: with a 80 ms RTT the maximum transfer speed is about 6.5 Mbps
with the current default buffer values
(https://www.switch.ch/network/tools/tcp_throughput/)
I've done some testing and auto tuning seems to be working as expected.
When downloading from another user (Windows -> Linux) with a 20 ms RTT
and buffer sizes set to 0 in both clients, I was able to utilize my 100
Mbps connection with a single thread, while using the old buffer size
limited the speed to ~30 Mbps, which matches the maximum speed given by
the calculator.
The required code changes are quite trivial: https://github.com/airdcpp
/airdcpp-windows/commit/f96ef123c8152860bb796208cadbe051cc5a8da8
** Affects: dcplusplus
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of
Dcplusplus-team, which is subscribed to DC++.
https://bugs.launchpad.net/bugs/1909853
Title:
Don't use custom socket buffer sizes by default
Status in DC++:
New
Bug description:
Modern operating systems (Linux 2.4+, Windows Vista+) should support
TCP tuning (https://en.wikipedia.org/wiki/TCP_tuning) to enable higher
per-thread speeds with fast connections. However, the current default
socket buffer size (65536 bytes) set by the client will prevent the
feature from being utilized.
Example: with a 80 ms RTT the maximum transfer speed is about 6.5 Mbps
with the current default buffer values
(https://www.switch.ch/network/tools/tcp_throughput/)
I've done some testing and auto tuning seems to be working as
expected. When downloading from another user (Windows -> Linux) with a
20 ms RTT and buffer sizes set to 0 in both clients, I was able to
utilize my 100 Mbps connection with a single thread, while using the
old buffer size limited the speed to ~30 Mbps, which matches the
maximum speed given by the calculator.
The required code changes are quite trivial:
https://github.com/airdcpp/airdcpp-
windows/commit/f96ef123c8152860bb796208cadbe051cc5a8da8
To manage notifications about this bug go to:
https://bugs.launchpad.net/dcplusplus/+bug/1909853/+subscriptions
Follow ups