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

[Andrea Bittau]

CVSROOT:	/usr/local/www/data/cvs
Module name:	xorp
Changes by:	abittau at chum.icir.org	2008-08-06 08:08:31 UTC

XORP CVS repository


Modified files:
	policy        visitor_semantic.cc 
	policy/backend iv_exec.cc 
	policy/common register_operations.cc 

Log message:
	Don't create temporary objects when executing.  (Only for the AsPath prepend
	case for now).  The basic problem is as follows.  Say you have an expression:
	
	1 + 5 + 10
	
	Before, I'd do 1 + 5 and create a new 6 object, and then 6 + 10, and create a 16
	object.  Now I do 1 + 5 and return the same 1 object but modified to 6.  I.e., I
	reuse one of the args to the expression.  It's a bit like using an accumulator
	argument when recursing I guess...
	
	This makes the interface much nastier - hence why I hate optimizing =D.
	
	Before:
	
	[58% 02:14 shorty test]# ./policybench -p base -v base.var  -i 100000
	Loading...
	Benchmarking.  Iterations: 100000
	Stats:
	Total time 1006 (ms) iterations 100000
	Iterations/s 99403.579
	Instr  1 Avg Time    182.589    ( 2.05%)
	Instr  2 Avg Time    653.278    ( 7.33%)
	Instr  3 Avg Time    850.118    ( 9.54%)
	Instr  4 Avg Time    311.957    ( 3.50%)
	Instr  5 Avg Time    657.300    ( 7.38%)
	Instr  6 Avg Time    222.927    ( 2.50%)
	Instr  7 Avg Time   2735.710    (30.71%)
	Instr  8 Avg Time    589.444    ( 6.62%)
	Instr  9 Avg Time    222.574    ( 2.50%)
	Instr 10 Avg Time   2481.870    (27.86%)
	[58% 01:58 shorty test]#
	
	Now:
	
	[47% 02:08 shorty test]# ./policybench -p base -v base.var  -i 100000
	Loading...
	Benchmarking.  Iterations: 100000
	Stats:
	Total time 838 (ms) iterations 100000
	Iterations/s 119331.742
	Instr  1 Avg Time    184.390    ( 2.54%)
	Instr  2 Avg Time    640.111    ( 8.83%)
	Instr  3 Avg Time    769.805    (10.62%)
	Instr  4 Avg Time    295.817    ( 4.08%)
	Instr  5 Avg Time    701.665    ( 9.68%)
	Instr  6 Avg Time    213.541    ( 2.95%)
	Instr  7 Avg Time   1172.950    (16.18%)
	Instr  8 Avg Time    609.467    ( 8.41%)
	Instr  9 Avg Time    215.179    ( 2.97%)
	Instr 10 Avg Time   2446.334    (33.75%)
	[47% 01:40 shorty test]#
	
	Look at Instr 7 - it's twice as fast.  I think BGP ASPath prepend is almost as
	optimized as it can get, at least for now.  I'll move on.  (Actually it seems
	like if Ive merely moved the problem eslewhere - instr 10 ;D.)

Revision  Changes                               Path
1.20      +16 -5;  commitid: 830448995be841a7;  xorp/policy/backend/iv_exec.cc
1.24      +5 -5;  commitid: 830448995be841a7;   xorp/policy/common/register_operations.cc
1.17      +11 -5;  commitid: 830448995be841a7;  xorp/policy/visitor_semantic.cc