Menu
▾
▴
[uml-devel] RE: [uml-user] [PATCH net-tools] debug improvements
|
From: Mooneer S. <mo...@ea...> - 2000-12-06 03:11:37
|
Kernel parameters: mem=64m no-xterm fake_arch
Host kernel: 2.2.17
UML kernel: 2.4.0-test11
Flood pinged the UML kernel with 1,000 ICMP packets and kept track of
packet loss/ping time. This was performed three times, with the best
result for each used.
"Traditional Style" (TCP to localhost):
root@apoc [~]# ping -c 1000 -f 10.0.0.1
PING 10.0.0.1 (10.0.0.1) from 10.0.0.2 : 56(84) bytes of data.
..........................................................................
--- 10.0.0.1 ping statistics ---
1073 packets transmitted, 1000 packets received, 6% packet loss
round-trip min/avg/max = 0.8/29.1/110.0 ms
root@apoc [~]#
Unix socket (/tmp/tap0):
root@apoc [~]# ping -c 1000 -f 10.0.0.1
PING 10.0.0.1 (10.0.0.1) from 10.0.0.2 : 56(84) bytes of data.
.
--- 10.0.0.1 ping statistics ---
1000 packets transmitted, 1000 packets received, 0% packet loss
round-trip min/avg/max = 0.3/0.3/1.8 ms
root@apoc [~]#
Table of Results (times in ms):
TCP Unix Difference
-----------------------------
Min| 0.8 0.3 0.5
Max| 29.1 0.3 28.8
Avg| 110.0 1.8 108.2
Pak| 6% 0% 0%
-----------------------------
Pak = Packet loss (percentage)
Min = Minimum ping time
Max = Maximum ping time
Avg = Average ping time
TCP = UML communicates with um_eth_net_util through a TCP/IP connection
Unix = UML communicates with um_eth_net_util through a named pipe/Unix
socket
I believe the reason for the speed increase is because the kernel
does not have to go through the overhead that a TCP/IP connection
entails, even when using the loopback device.
-----Original Message-----
From: Jeff Dike [mailto:jd...@ka...]
Sent: Tuesday, December 05, 2000 5:27 PM
To: Mooneer Salem
Cc: Rodrigo Barbosa (aka morcego);
use...@li...
Subject: Re: [uml-user] [PATCH net-tools] debug improvements
rod...@co... said:
> Have you benckmarked it ?
Rodrigo is right. Measure the latency across both types of sockets. A
connection through localhost goes through the loopback interface, which I
think is about as low-overhead as you could want.
Jeff
|