From mboxrd@z Thu Jan 1 00:00:00 1970 From: bogus@does.not.exist.com () Date: Mon, 4 May 2009 19:42:15 +0000 Subject: No subject Message-ID: Topicbox-Message-UUID: 005a861e-ead5-11e9-9d60-3106f5b1d025 "4.1 Visualizing trace device output Once we had the data, we needed a way to analyse the information. After working with the data for a while, we realized that the output as shown in Figure 1 would be very useful. No graphiing [sic] tool available to us in Plan 9 or Linux was able to create that output. In the end, we determined that gnuplot was the most appropriate tool, but even then the data required significant processing to get it into the proper form. We wrote a suite of scripts usng rc, the plan 9 shell; acid, the Plan 9 debugger; awk, and sed to generate data appropriate for plotting with gnuplot. The createplot script has the ability to filter out functions which ran for less than a specified number of clock cycles, which is useful for reducing the amount of noise in a plot. To generate a plot from the data collected earlier, discarding functions which completed in less than 4000 cycles, we just ran: plots/createplot /amd64/9k8pf 4000 ./trace > plotme and fed the input into gnuplot." Jason Catena