netdev - low TCP throughput for small MTU and high delay

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Date:	Fri, 6 Apr 2007 15:42:54 +0800
From:	"Way Chuang Ang" <wcang79@...il.com>
To:	netdev@...r.kernel.org
Subject: low TCP throughput for small MTU and high delay

Hi all,

I encountered a strange behaviour during iperf TCP test. When I set
MTU of ethernet device to 170 and incurred a delay of 250ms via netem
on sender's side, the throughput was 115Kbps. From the output of
tcp_probe the snd_cwnd peaked and stayed at 50 for the whole duration
of the test and snd_ssthresh was 100. It never went pass slow start.
>From the output of the sender is always hindered by the small window
size of 5824 bytes advertised by the receiver.

Under the same situation where all the environment variables were
same, I set MTU of sender to 180 and repeated the iperf test. It
managed to achieve average throughput of 2.18Mbps. snd_cwnd (1252)
managed to exceed snd_ssthresh (1117). The weird thing is the receiver
advertised a bigger window size of 371968, thus sender can send more
data before waiting for acknowledgement.

So my question is what causes the receiver to advertise bigger window
size for 180 byte MTU? If you have any explanation for this behaviour,
please tell. The sender advertised MSS of 256bytes for both MTU sizes.
I tried TCP hybla as well, but the throughput results were the same
with the default (BIC).

Both sender (x86 - 32 bit) and receiver  (x86- 64bit) are Fedora Core
6 machines. Brief specs of these machines:
Distro: FC6
Kernel version: 2.6.18-1.2849
Processor: Intel P4 2.8 GHz

Thank you in advance.

Regards,
Ang Way Chuang
--------------------------
aka Idiot
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html