Date: Sun, 28 Sep 2008 11:06:13 -0700 From: "Garrett Cooper" <yanefbsd@gmail.com> To: "William LeFebvre" <bill@lefebvre.org> Cc: Alex Keda <admin@lissyara.su>, Edwin Groothuis <edwin@mavetju.org>, stable@freebsd.org, current@freebsd.org Subject: Re: Request for testing - top 3.8b1 in the base system Message-ID: <7d6fde3d0809281106o331a495cg8d5c3c65c521af12@mail.gmail.com> In-Reply-To: <48DFADF6.10504@lefebvre.org> References: <20080928054620.GA80250@k7.mavetju> <48DF3CFE.7@lissyara.su> <7d6fde3d0809280209i3003829bj23baa93f0b271163@mail.gmail.com> <20080928112538.GC13745@k7.mavetju> <48DFADF6.10504@lefebvre.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, Sep 28, 2008 at 9:16 AM, William LeFebvre <bill@lefebvre.org> wrote: > Edwin Groothuis wrote: >> >> On Sun, Sep 28, 2008 at 02:09:00AM -0700, Garrett Cooper wrote: >>> >>> On Sun, Sep 28, 2008 at 1:14 AM, Alex Keda <admin@lissyara.su> wrote: >>>> >>>> Some strange. Count running processes not match with system top >> >> That has been explained in an email before. >> >>> I'm not sure I'm finding an issue, but I do find it interesting that... >>> 1. It takes a reasonably long amount of time for top to plateau the >>> WCPU field (approximately 8-10 iterations), whereas ps registering the >>> WCPU percentage value is almost instantaneous. > > Top 3.8 doesn't display WCPU. It is an antequated measure that is only > maintained by the kernel so that ps can display it. It no longer has any > meaning to the scheduler, so why bother displaying it. > >> >> With ps it takes 10 2 second steps to get the WCPU from 0 to 100, >> with the new top (which doesn't have WCPU (See Changes file, and >> the m_freebsd.c file, I don't know of the real reason behind it) >> anymore) goes from 0 to 100 in 2 2 second steps. > > ps shows a decaying average as calculated by the kernel over the past minute > and recorded in the proc structure. Top calculates its own average based on > the difference in cpu time between the last measurement and the current > measurement. The output from ps is fine when you want a single snapshot: > you want it to show information averaged over a long period of time. Top is > showing you only what's going on right now, since the last update. That's > why percent CPU in top will climb to its final value so quickly. > > Bill LeFebvre Actually, I was trying to say it was the other way around -- WCPU took a long time in top to climb to its final value where it took a short period of time with ps. Retrying that though, it appears that I was flip-flopping my statement and yes it aligns with Bill's. I still find the averaging discrepancy a bit interesting, but it's merely a function of how the average is being taken. -Garrett
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?7d6fde3d0809281106o331a495cg8d5c3c65c521af12>