problem: performance gets roughly halved on a single processor box when
CONFIG_X86_UP_IOAPIC is enabled.
hardware: compaq deskpro. 400MHz PII/64Mb RAM/Intel PIIX4 controller, eepro100
network card.
Enabling APIC on single processor boxes seems to roughly halve the performance:
this isn't benchmark rates, but real world applications, e.g. the following
from /usr/bin/time:
2.3.42 without APIC
Command being timed: "bunzip2 tmp.tar.bz2"
User time (seconds): 17.86
System time (seconds): 0.21
Percent of CPU this job got: 99%
Elapsed (wall clock) time (h:mm:ss or m:ss): 0:18.09
Major (requiring I/O) page faults: 102
Minor (reclaiming a frame) page faults: 911
Page size (bytes): 4096
2.3.42 with APIC Command being timed: "bunzip2 tmp.tar.bz2"
User time (seconds): 37.80
System time (seconds): 0.42
Percent of CPU this job got: 196%
Elapsed (wall clock) time (h:mm:ss or m:ss): 0:19.49
Major (requiring I/O) page faults: 534
Minor (reclaiming a frame) page faults: 911
(bits snipped that didn't change). Note that the number of major page faults
has jumped massively (and the wall clock time was completely wrong). I've also
noticed that /proc/interrupts shows that the number of interrupts on IRQ 11
(usb-uhci & ethernet) seems much higher with APIC enabled.
Can't see any immediate fix, apart from a change to the config menu to put a
warning and maybe remove the comment "Such a kernel will still boot on
IO-APIC-less sstems with no slowdown at all."
I've got a feeling this is more to do with educating people like me who tend to
select 'y' for anything they've not tried before than actually being a bug but
it's probably worth clarifying.
cheers
john
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Feb 23 2000 - 21:00:28 EST