|
I am currently trying to tune LVS. I ran machines (everyone running
Apache 1.3.20) behind a LVS (NAT, round-robin/wlc) machine. I made some
benchmarks with a wrapper around apachebench that varys the size of the
payload,
number of threads when running against the LVS.
Here are the results:
100 ; 25 ; 580.43 ; 196.19
100 ; 50 ; 527.47 ; 178.28
100 ; 100 ; 513.75 ; 173.65
100 ; 250 ; 587.47 ; 198.56
1000 ; 25 ; 770.17 ; 938.68
1000 ; 50 ; 740.39 ; 902.38
1000 ; 100 ; 713.44 ; 869.55
1000 ; 250 ; 714.02 ; 870.25
10000 ; 25 ; 323.02 ; 3233.36
10000 ; 50 ; 345.41 ; 3457.47
10000 ; 100 ; 308.51 ; 3088.09
10000 ; 250 ; 287.17 ; 2874.44
25000 ; 25 ; 164.03 ; 4044.59
25000 ; 50 ; 163.07 ; 4021.18
25000 ; 100 ; 198.44 ; 4892.99
50000 ; 25 ; 108.25 ; 5311.72
50000 ; 50 ; 128.76 ; 6318.36
100000 ; 25 ; 15.38 ; 1505.62
100000 ; 50 ; 7.47 ; 731.93
Col 1: bytes of the requested document
Col 2: number of threads
Col 3: Requests/Second
Col 4: KB transfered/second
The testscript ran every test several times. So the numbers
are average numbers.
I wonder why the LVS only transfers about 800-900 KB/second when the
document size is 1.000 bytes but is able to transfer 6300kb/second
with a document size of 50.000 bytes. And why there a break in the
performance
with documents that are 100.000 bytes large ?
The LVS server is a PenIII with 733 MHz. The five servers are Pen-III/1GHz.
Andreas Jung
---------------------------------------------------------------------
- Andreas Jung Zope Corporation -
- EMail: andreas@xxxxxxxx http://www.zope.com -
- "Python Powered" http://www.python.org -
- "Makers of Zope" http://www.zope.org -
- "Endings are just Beginnings" -
---------------------------------------------------------------------
|
