Subject: lmbench pmax -G0 vs -G10 results
To: None <port-mips@netbsd.org>
From: Simon Burge <simonb@wasabisystems.com>
List: port-mips
Date: 04/04/2002 12:51:29
Here's the results of running lmbench on a pmax (5000/260, 60MHz R4400)
built with -G0 and -G10. Each line is really the average of three runs.
L M B E N C H 1 . 9 S U M M A R Y
------------------------------------
(Alpha software, do not distribute)
Processor, Processes - times in microseconds - smaller is better
----------------------------------------------------------------
Host OS Mhz null null open selct sig sig fork exec sh
call I/O stat clos inst hndl proc proc proc
--------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ---- ---- ----
pmax-G 0 NetBSD 1.5ZC 118 2.2 22. 113 159 0.60K 6.6 31 10.4K 54K 98K
pmax-G 10 NetBSD 1.5ZC 118 2.2 21. 101 124 0.39K 7.1 33 9.9K 53K 95K
Context switching - times in microseconds - smaller is better
-------------------------------------------------------------
Host OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw
--------- ------------- ----- ------ ------ ------ ------ ------- -------
pmax-G 0 NetBSD 1.5ZC 25 159 522 212 942 232 1506
pmax-G 10 NetBSD 1.5ZC 27 141 547 186 875 223 1452
*Local* Communication latencies in microseconds - smaller is better
-------------------------------------------------------------------
Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP
ctxsw UNIX UDP TCP conn
--------- ------------- ----- ----- ---- ----- ----- ----- ----- ----
pmax-G 0 NetBSD 1.5ZC 25 109 131 426 459 2422
pmax-G 10 NetBSD 1.5ZC 27 106 120 388 434 2030
File & VM system latencies in microseconds - smaller is better
--------------------------------------------------------------
Host OS 0K File 10K File Mmap Prot Page
Create Delete Create Delete Latency Fault Fault
--------- ------------- ------ ------ ------ ------ ------- ----- -----
pmax-G 0 NetBSD 1.5ZC 3497 1136 4485 4059 16644 3 9.1K
pmax-G 10 NetBSD 1.5ZC 3751 1145 4545 3813 19427 9.1K
*Local* Communication bandwidths in MB/s - bigger is better
-----------------------------------------------------------
Host OS Pipe AF TCP File Mmap Bcopy Bcopy Mem Mem
UNIX reread reread (libc) (hand) read write
--------- ------------- ---- ---- ---- ------ ------ ------ ------ ---- -----
pmax-G 0 NetBSD 1.5ZC 16 19 9 9 24 10 10 24 18
pmax-G 10 NetBSD 1.5ZC 15 19 9 9 24 10 10 24 18
Memory latencies in nanoseconds - smaller is better
(WARNING - may not be correct, check graphs)
---------------------------------------------------
Host OS Mhz L1 $ L2 $ Main mem Guesses
--------- ------------- --- ---- ---- -------- -------
pmax-G 0 NetBSD 1.5ZC 118 25 126 1252
pmax-G 10 NetBSD 1.5ZC 118 25 126 1253
stat, open/close and select are all reasonably better with -G 10.
Everything else seems reasonably line ball.
The kernel sizes are:
text data bss dec hex filename
2490008 56800 432228 2979036 2d74dc netbsd.G0
2434800 76224 431044 2942068 2ce474 netbsd.G10
The -G10 kernel is ~1% smaller.
So, a slight performance win in a couple of cases and slightly smaller
kernels for the -G 10 case. I'm almost tempted to say we should use
-G 8 or -G 10 on all mips ports.
Comments?
Simon.
--
Simon Burge <simonb@wasabisystems.com>
NetBSD CDs, Support and Service: http://www.wasabisystems.com/