Package: sysutils
Version: 2.0.1

>From the description:
   bogomips will be just dropped, use "cat /proc/cpuinfo | grep ^bogo" instead.

This is not equivalent to the userspace bogomips tool.

The standalone bogomips program offered several differences from the
kernel bogomips:
 - showed a number based on current processor settings
 -- this is useful to quickly gauge how much of a performance hit is
incurred by changing processor speed, throttling, etc
 - showed a number which is affected by current running processes
 -- this can be useful to immediately gauge the effects of "nice" and
similar methods, as well as killing off processes
 - does not show a static number as calculated at boot time
 -- /proc/cpuinfo does not update the bogomips field, as near as I can
tell.  It does update the cpu_mhz field, but this does not include the
effect of processor throttling.  (Some kernels /do/ update
/proc/cpuinfo bogomips:, but they merely scale the originally
calculated value by the current processor clock.)


I suggest that a bogomips package be created, incorporating the
standalone version of the bogomips utility.  It may be desirable to
enhance the utility; offering both versions of the current bogomips
number, perhaps by command line switches.

Realistically, there are several bogomips numbers that a user might be
interested in:
 - the number originally calculated at boot time (likely with full
processor speed and no throttling - "dmesg | grep Bogo")
 - the number currently employed by the kernel (scaled, according to
current processor speed - "cat /proc/cpuinfo | grep ^bogo")
 - the number currently employed by the kernel (scaled according to
processor speed and throttling)
 - the number currently available in userspace

It may also be useful to disable processor speed changes during the
calculation - running "nice" may be adequate for this, as several
processor speed utilities do not turn up processor speed for the
benefit of "nice" processes.



-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to