From mboxrd@z Thu Jan 1 00:00:00 1970 From: erik quanstrom Date: Mon, 5 Mar 2012 11:14:05 -0500 To: 9fans@hamnavoe.com, 9fans@9fans.net Message-ID: In-Reply-To: References: MIME-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Subject: Re: [9fans] usb slowness Topicbox-Message-UUID: 66a10f54-ead7-11e9-9d60-3106f5b1d025 On Mon Mar 5 11:06:43 EST 2012, 9fans@hamnavoe.com wrote: > > if you could temporarly run a 9atom kernel, >=20 > With the normal kernel, you can use acid or db to look at the intrtimes > array to see the spread of cpu times used in each interrupt vector entr= y. >=20 > See http://9fans.net/archive/2003/11/520 i didn't find the original histogram very useful. it is not scaled by the speed of the processor nor type of interrupt. the slots in the histogram are appropriate for 8259 interrupts on a 2000-era processor (100=C2=B5s increments). almost all the interrupts i see are < 20=C2=B5s. the 82598 on my relativly slow cpu server averages 2.5=C2=B5= s per interrupt. it's also a pain to print out without a file in addition to /dev/irqalloc= . i found the absolute count and absolute # of cpu cycles (to avoid predivision sloppiness) to be a lot more useful for the general case. a hisogram sure could be useful, but i think i'd want to taylor it to a specific problem. and i haven't had one yet. - erik