Subject: VNODE_OP_NOINLINE tests on pmax (5000/260).
To: None <tech-kern@netbsd.org>
From: Simon Burge <simonb@wasabisystems.com>
List: tech-kern
Date: 09/15/2000 12:21:48
Here's some lmbench runs without and with VNODE_OP_NOINLINE.
"pmax-before" doesn't have the option set, and "pmax-after"
does have it set. The later kernel is just a tad over 37k
smaller.
There are two runs of each, one after the other, just after
a reboot in each case. Some benchmarks are a bit quicker,
some a bit slower but overall no obvious difference.
Also, some times to do a "make" in my kernel compile directory.
The sequence used was:
cd <build-dir>
make clean
copy "before" kernel to /
reboot
cd <build-dir>
time make
2470.516u 326.525s 51:32.16 90.4% 0+0k 4023+20548io 1124pf+0w
make clean
copy "after" kernel to /
reboot
cd <build-dir>
time make
2435.594u 326.638s 50:57.29 90.3% 0+0k 4013+20429io 1124pf+0w
So for the "make kernel test" it's ~1% faster with VNODE_OP_NOINLINE.
Cheers,
Simon.
--
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-before NetBSD 1.5E 118 3.4 31. 114 155 0.32K 9.2 30 4.6K 40K 62K
pmax-before NetBSD 1.5E 118 3.4 31. 113 154 0.32K 9.2 30 4.9K 40K 62K
pmax-after NetBSD 1.5E 118 3.2 24. 104 126 0.37K 9.7 31 4.7K 40K 63K
pmax-after NetBSD 1.5E 118 3.2 23. 103 125 0.37K 9.3 32 4.7K 40K 63K
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-before NetBSD 1.5E 20 194 496 216 592 207 1280
pmax-before NetBSD 1.5E 17 132 558 156 1239 188 1619
pmax-after NetBSD 1.5E 16 150 434 255 848 249 1279
pmax-after NetBSD 1.5E 15 198 528 173 1422 283 1769
*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-before NetBSD 1.5E 20 127 126
pmax-before NetBSD 1.5E 17 127 126
pmax-after NetBSD 1.5E 16 156 112
pmax-after NetBSD 1.5E 15 114 114
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-before NetBSD 1.5E 2631 1282 4000 4545 154862 1 5.8K
pmax-before NetBSD 1.5E 2941 1136 5263 2941 152840 1 4.5K
pmax-after NetBSD 1.5E 2777 1136 3448 4761 156743 2 5.2K
pmax-after NetBSD 1.5E 2702 1136 3333 4545 158504 1 5.1K
*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-before NetBSD 1.5E 18 18 -1 9 24 10 10 24 18
pmax-before NetBSD 1.5E 19 18 -1 9 24 10 10 24 18
pmax-after NetBSD 1.5E 19 19 -1 8 24 10 10 24 18
pmax-after NetBSD 1.5E 16 14 -1 8 24 10 9 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-before NetBSD 1.5E 118 25 126 1246
pmax-before NetBSD 1.5E 118 25 126 1247
pmax-after NetBSD 1.5E 118 25 125 1246
pmax-after NetBSD 1.5E 118 25 199 1248
--
Simon Burge <simonb@wasabisystems.com>
NetBSD Sales, Support and Service: http://www.wasabisystems.com/