Hardware virtualization slower than software?
http://developers.slashdot.org/developers/06/08/12/2028223.shtml
... from above:
One example given is compilation of a Linux kernel under a virtualized
Linux OS. Native wall-clock time: 265 seconds. Software-assisted
virtualization: 393 seconds. Hardware-assisted virtualization: 484
seconds. Ouch. It sounds to me like a hybrid approach may be the best
answer to the virtualization problem. "
... snip ...
similar, but different posting made here not too long ago
http://www.garlic.com/~lynn/2006j.html#27 virtual memory
the above has a discussion about hardware/software virtualization
trade-off in 3081 vis-a-vis 3090. note that this was pre-"PR/SM"
(which has since evolved into LPARS) ... where the microcode support
can create virtual machines ... w/o requiring separate hypervisor
monitor running (i.e. "dropping" everything into hardware was no
longer a performance trade-off issue).
for other drift, the performance characterization in the article is
reminisent of presentation i made at boston share meeting in aug68.
three people had came out from the science center
http://www.garlic.com/~lynn/subtopic.html#545tech
the last week in jan68 to install cp67 at the university. during the
spring and summer of 68, i rewrote significant poritions of the
kernel, some cases descreasing pathlengths by factor of 10 to 100
times. past posting of parts of that presentation
http://www.garlic.com/~lynn/94.html#18 CP/67 and OS MFT14
which has a bare-machine (native wall clock) time of 322 sec.
original virtualization elapsed time 856 sec. virtualization elapsed
time (after rewrites of the spring and summer) 435 secs
(virtualization processing was reduced from 534 cpu secs. to 113 cpu
secs.).
