[Xorp-cvs] XORP cvs commit: xorp/policy/backend xorp/policy/common xorp/policy/test

Andrea Bittau abittau at icir.org
Wed Aug 6 01:11:45 PDT 2008


CVSROOT:	/usr/local/www/data/cvs
Module name:	xorp
Changes by:	abittau at chum.icir.org	2008-08-06 08:11:45 UTC

XORP CVS repository


Modified files:
	policy/backend iv_exec.cc 
	policy/common register_operations.cc 
	policy/test   Jamfile Makefile.am Makefile.in file_varrw.cc 
	              policybench.cc policyvarmap.conf 

Log message:
	Added BGPVarRW to policybench.  This allows measuring the policy performance in
	BGP's context.
	
	I've backed out the optimization of ASPaths since it's not compatible with BGP's
	attribute manager.  (The manager keeps only one copy, so if we recycle, we alter
	the only-one shared copy.)  We've lost 22% performance due to this, but I'll try
	to get it back somehow.
	
	The current situation:
	With FileVarRW (essentially the performance of filters themselves):
	[27% 01:03 shorty test]# ./policybench -p base -t 0 -v base.var
	Loading...
	Benchmarking.  Iterations: 100000
	Stats:
	Total time 457 (ms) iterations 100000
	Iterations/s 218818.381
	Instr  1 Avg Time    182.261    ( 3.87%)
	Instr  2 Avg Time    404.978    ( 8.60%)
	Instr  3 Avg Time    474.253    (10.07%)
	Instr  4 Avg Time    260.620    ( 5.53%)
	Instr  5 Avg Time    243.838    ( 5.18%)
	Instr  6 Avg Time    210.663    ( 4.47%)
	Instr  7 Avg Time   2242.036    (47.60%)
	Instr  8 Avg Time    240.580    ( 5.11%)
	Instr  9 Avg Time    211.268    ( 4.49%)
	Instr 10 Avg Time    239.414    ( 5.08%)
	
	With BGPVarRW:
	[27% 01:00 shorty test]# ./policybench -p base -t 1
	Loading...
	Benchmarking.  Iterations: 100000
	Stats:
	Total time 2942 (ms) iterations 100000
	Iterations/s 33990.483
	Instr  1 Avg Time    263.791    ( 3.83%)
	Instr  2 Avg Time    871.429    (12.66%)
	Instr  3 Avg Time    863.823    (12.55%)
	Instr  4 Avg Time    372.550    ( 5.41%)
	Instr  5 Avg Time    431.017    ( 6.26%)
	Instr  6 Avg Time    211.631    ( 3.07%)
	Instr  7 Avg Time   2971.534    (43.17%)
	Instr  8 Avg Time    427.506    ( 6.21%)
	Instr  9 Avg Time    211.391    ( 3.07%)
	Instr 10 Avg Time    258.085    ( 3.75%)
	
	BGP's varrw is 6x slower than an "optimal" implementation.  The goal for now is
	to speed up BGP's varrw by 3x.

Revision  Changes                                 Path
1.24      +3 -2;  commitid: 84ac48995cbd41a7;     xorp/policy/backend/iv_exec.cc
1.27      +5 -9;  commitid: 84ac48995cbd41a7;     xorp/policy/common/register_operations.cc
1.4       +13 -1;  commitid: 84ac48995cbd41a7;    xorp/policy/test/Jamfile
1.7       +15 -4;  commitid: 84ac48995cbd41a7;    xorp/policy/test/Makefile.am
1.22      +26 -4;  commitid: 84ac48995cbd41a7;    xorp/policy/test/Makefile.in
1.17      +2 -2;  commitid: 84ac48995cbd41a7;     xorp/policy/test/file_varrw.cc
1.5       +124 -29;  commitid: 84ac48995cbd41a7;  xorp/policy/test/policybench.cc
1.6       +14 -29;  commitid: 84ac48995cbd41a7;   xorp/policy/test/policyvarmap.conf



More information about the Xorp-cvs mailing list