Difference between revisions of "Network Tuning"

Latest revision as of 21:33, 24 November 2010

Converted with HTML::WikiConverter::MediaWiki from my old phpwiki site

Links

Bridges & Proxy-ARP

Traffic shaping

Tools

Iperf

Examples of a WET11 to WET11 communication:

apt-get install iperf

Server: (TCP & UDP via default port 5001)

iperf -s -D
iperf -s -D -u

Client:

 iperf -c 10.0.0.9
 ------------------------------------------------------------
 Client connecting to 10.0.0.9, TCP port 5001
 TCP window size: 16.0 KByte (default)
 ------------------------------------------------------------
 [  5] local 10.0.0.136 port 1089 connected with 10.0.0.9 port 5001
 [ ID] Interval       Transfer     Bandwidth
 [  5]  0.0-10.1 sec  6.41 MBytes  5.30 Mbits/sec
 
 iperf -c 10.0.0.9 -u -b 100M
 ------------------------------------------------------------
 Client connecting to 10.0.0.9, UDP port 5001
 Sending 1470 byte datagrams
 UDP buffer size: 64.0 KByte (default)
 ------------------------------------------------------------
 [  5] local 10.0.0.136 port 1024 connected with 10.0.0.9 port 5001
 [ ID] Interval       Transfer     Bandwidth
 [  5]  0.0-10.0 sec  11.4 MBytes  9.59 Mbits/sec
 [  5] Server Report:
 [  5]  0.0-10.1 sec  7.70 MBytes  6.37 Mbits/sec  3.415 ms 2668/ 8157 (33%)
 [  5] Sent 8157 datagrams

Interesting test:

 iperf -c 127.0.0.1
 ------------------------------------------------------------
 Client connecting to 127.0.0.1, TCP port 5001
 TCP window size: 49.3 KByte (default)
 ------------------------------------------------------------
 [  3] local 127.0.0.1 port 51547 connected with 127.0.0.1 port 5001
 [  3]  0.0-10.0 sec  3.99 GBytes  3.43 Gbits/sec
 
 iperf -c 127.0.0.1 -u -b 1G
 ------------------------------------------------------------
 Client connecting to 127.0.0.1, UDP port 5001
 Sending 1470 byte datagrams
 UDP buffer size:   103 KByte (default)
 ------------------------------------------------------------
 [  3] local 127.0.0.1 port 32769 connected with 127.0.0.1 port 5001
 [  3]  0.0-10.0 sec    941 MBytes    789 Mbits/sec
 [  3] Sent 671249 datagrams
 [  3] Server Report:
 [  3]  0.0-10.0 sec    934 MBytes    785 Mbits/sec  0.000 ms 4831/671248 (0.72%)
 [  3]  0.0-10.0 sec  1 datagrams received out-of-order

Looks like the kernel handles UDP much worsely than TCP!!
( tests done with a 2.6.15-1-686 )

4150 / mins = K/s

apt-get install iproute

Tests de ping sur www.ulg.ac.be

NORMAL
round-trip min/avg/max = 13.9/18.3/23.9 ms

Tests avec download (wget -O /dev/null http://ftp.belnet.be/debian-cd/images/current/i386/debian-30r1-i386-binary-1.iso)

DOWNLINK=2000 THEORETIC LIMIT
Download 1600 kbps
round-trip min/avg/max = 20.8/71.6/152.3 ms
DOWNLINK=800
Download 584 kbps
round-trip min/avg/max = 15.8/27.7/58.0 ms
DOWNLINK=1500
Download 1160 kbps
round-trip min/avg/max = 14.8/46.9/142.8 ms
DOWNLINK=1600
Download 1160 kbps
round-trip min/avg/max = 20.8/46.7/118.0 ms
DOWNLINK=2000
Download 1320 kbps
round-trip min/avg/max = 18.9/64.0/157.5 ms

Tests avec upload (scp /home/server/www/win98-v02.img dorian.homelinux.org:/dev/null)

UPLINK=512 THEORETIC LIMIT
Upload 416.0 kbps
round-trip min/avg/max = 18.9/138.4/258.1 ms
UPLINK=220
Upload 207.2 kbps
round-trip min/avg/max = 14.5/38.7/72.4 ms
UPLINK=440
Upload 425.6 kbps
round-trip min/avg/max = 40.4/147.7/317.9 ms
UPLINK=400
Upload 381.6 kbps
round-trip min/avg/max = 14.9/43.5/80.3 ms

Bandwidth analysis

Ipband

Darkstat

Backport sur la Woody:

apt-get source darkstat (sur une Sarge) -> transfert sur la Woody
cd darkstat-2.5
apt-get install autotools-dev libpcap-dev
debian/control l10: debconf (>=1.2.9) -> (>=1.0.32)
debian/rules binary
cd ..
dpkg -i darkstat_2.5-4_i386.deb
editer /etc/darkstat/init.cfg
/etc/init.d/darkstat start

Autres

iptraf
argus
ntop