Path: csiph.com!x330-a1.tempe.blueboxinc.net!usenet.pasdenom.info!weretis.net!feeder4.news.weretis.net!news.musoftware.de!wum.musoftware.de!fu-berlin.de!uni-berlin.de!individual.net!not-for-mail
From: Rainer Weikusat <rweikusat@mssgmbh.com>
Newsgroups: comp.os.linux.development.system
Subject: Re: How to estimate System Memory Bandwidth by sampling
Date: Fri, 15 Jul 2011 16:24:31 +0100
Lines: 23
Message-ID: <87y5zzin2o.fsf@sapphire.mobileactivedefense.com>
References: <cff4b142-7330-4913-b122-b69b4b430d28@s33g2000prg.googlegroups.com> <92gq1790rij00c57o68guvvjjpmpa9qmu0@4ax.com> <87mxgi4bxk.fsf@sapphire.mobileactivedefense.com> <96jv175030s6dgs36egstrgvg9bi8mk9rf@4ax.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Trace: individual.net zOUxgFSdwZ0azoewaxcjXg6HvYQ9XGMHo08j6ySkTTrwnAxXM=
Cancel-Lock: sha1:W64tbDqRou1k2HoixDFjyPrZLMs= sha1:f6Mq6KR8YZYEb4Jcwxw7LlrWoFw=
User-Agent: Gnus/5.11 (Gnus v5.11) Emacs/22.2 (gnu/linux)
Xref: x330-a1.tempe.blueboxinc.net comp.os.linux.development.system:218

Tim Roberts <timr@probo.com> writes:
> Rainer Weikusat <rweikusat@mssgmbh.com> wrote:
>>Why this? All which seems to be needed for that is access to an area
>>of physical memory configured as uncached and some hardware
>>timer that can be used to measure the time it takes to execute some
>>'block memory access loop'.
>
> Assuming you stopped all of the other CPUs and blocked any peripherals that
> might do DMA, you could compute the bandwidth of the path between CPU and
> memory.

Interrupts would need to be disabled, too. But in practice, that's
easier to do than it seems: Take the system into 'single user mode'
and disconnect any kind of network. This should at least result in a
useful approximation.

[...]

> More importantly, however, there is still no way to find out "how much
> bandwidth was used in the last N milliseconds".  That's what he wanted to
> find out.

Ok, I misread that ...