iedowse> Someone mentioned on a list somewhere that vmware takes forever to
iedowse> emulate the cmpxchg instruction, and that using the I386_CPU version
iedowse> of atomic_cmpset_int() helps a lot.
I really know I'm doing a stupid thing, but here is benchmark results
of both "plain" and "patched" 5-current (as of Nov/26/2001). Patched
FreeBSD is about 10% faster than before.
*** Before:
TEST BASELINE RESULT INDEX
Arithmetic Test (type = double) 2541.7 156596.8 61.6
Dhrystone 2 without register variables 22366.3 1214131.7 54.3
Execl Throughput Test 16.5 25.1 1.5
File Copy (30 seconds) 179.0 1684.0 9.4
Pipe-based Context Switching Test 1318.5 710.9 0.5
Shell scripts (8 concurrent) 4.0 7.0 1.8
=========
SUM of 6 items 129.1
AVERAGE 21.5
*** After:
TEST BASELINE RESULT INDEX
Arithmetic Test (type = double) 2541.7 167038.3 65.7
Dhrystone 2 without register variables 22366.3 1267100.0 56.7
Execl Throughput Test 16.5 45.0 2.7
File Copy (30 seconds) 179.0 2863.0 16.0
Pipe-based Context Switching Test 1318.5 1372.6 1.0
Shell scripts (8 concurrent) 4.0 12.5 3.1
=========
SUM of 6 items 145.3
AVERAGE 24.2
***
Note that both are tested with:
- Same kernel configuration (but not GENERIC kernel)
- VMware Workstation 3.0.0 build 1455, WindowsXP Pro host
- 96MB RAM for FreeBSD guest OS
- 1.9GB Virtual Disk is on ATA66 HDD of host PC.
- Host PC has one Pentium3 850Mhz CPU
-- -
Makoto `MAR' Matsushita
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-hackers" in the body of the message
