Skip to content

Optimal TCP Congestion Control for 1G Ethernet

July 18, 2010

I was tuning a TCP/IP based system for Java in Linux to be used among almost 300 G6 servers. I thinks this is gono be the largest server farm currently deployed in Iran. What do we do with this enormous computational power? find criminals:) but among many other tuning elements inside OS and JVM, I also run a load test with iperf changing TCP congestion control.

I used Ubuntu 10.04 Server PAE kernel with increased rmem and wmem buffer space. Network in each rack is *1G*. Here is the results. I also found better results with No Delay and run all tests in this mode. MTU is 1500. Switch is Cisco Catalyst 2900.

The difference was so minimal. scalable due is the winner.

Algorithms 1 2 3 Avg (kb/s) Avg (MB/s) Performance
bic 932045 931186 931182 931471 113.704 0.9538
highspeed 931122 931036 931617 931258.33 113.678 0.9536
htcp 930762 931349 931855 931322 113.6867 0.9536
hybla 926258 926624 925711 926197.667 113.0612 0.9484
illinois 925670 926369 927478 926505.667 113.098 0.9487
lp 928534 925703 929403 927880 113.2666 0.9501
scalable 933914 934299 933616 933943 114.006 0.95635
vegas 699956 595055 612647 635886 77.6228 0.6511
veno 927823 931888 929281 929664 113.484 0.9519
westwood 930876 932491 926664 930010.334 113.5266 0.9523
yeah 925812 925480 927238 926176.667 113.0586 0.9483
Advertisements

From → linux, ubuntu

Leave a Comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: