https://homeostasis.scs.carleton.ca/wiki/index.php?action=history&feed=atom&title=Operating_Systems_2015F_Lecture_11 2026-05-01T05:38:15Z Revision history for this page on the wiki MediaWiki 1.42.1 https://homeostasis.scs.carleton.ca/wiki/index.php?title=Operating_Systems_2015F_Lecture_11&diff=20326&oldid=prev 2015-10-14T19:03:52Z

<p>Created page with &quot;==Video== The video for the lecture given on October 14, 2015 [http://homeostasis.scs.carleton.ca/~soma/os-2015f/lectures/comp3000-2015f-lec11-14Oct2015.mp4 is now available]...&quot;</p> <p><b>New page</b></p><div>==Video==<br /> <br /> The video for the lecture given on October 14, 2015 [http://homeostasis.scs.carleton.ca/~soma/os-2015f/lectures/comp3000-2015f-lec11-14Oct2015.mp4 is now available].<br /> <br /> ==Notes==<br /> <br /> &lt;pre&gt;<br /> Lecture 11<br /> ----------<br /> <br /> - ftrace<br /> - virtual memory<br /> <br /> Ftrace<br /> ------<br /> - as root, go into /sys/kernel/debug<br /> - enable events that you want by doing an echo &quot;1&quot; to the &quot;enable&quot; file for the event you want, e.g. events/syscalls/sys_enter_chdir/enable<br /> - look at the events in trace or, to just get the most recent ones, attach to trace_pipe<br /> - be sure to do an echo &quot;0&quot; to the enable file when you are done!<br /> <br /> For a tutorial on ftrace, see lwn.net:<br /> https://lwn.net/Articles/365835/ (Part 1)<br /> https://lwn.net/Articles/366796/ (Part 2)<br /> <br /> <br /> Virtual memory<br /> --------------<br /> - remember the kernel is LAZY<br /> - defers I/O for as long as possible, in general<br /> - &quot;mapped&quot; files are only loaded into memory logically at the time of the mmap<br /> - file I/O only happens when assocatied memory is accessed (read or write)<br /> - loads in blocks of course<br /> - data is loaded &quot;on demand&quot;<br /> <br /> - kernel is also lazy about memory allocation<br /> - consider fork<br /> - logically, copies a process&#039;s entire address space<br /> - often, though, that address space is quickly discarded<br /> (when you do an execve)<br /> - solution: only copy data *as needed*<br /> - called &quot;copy on write&quot;, or COW<br /> <br /> - kernel memory management is about using RAM to maximum effect<br /> - for files that are being currently accessed<br /> - only the part of those files that are being accessed<br /> - only code and data of processes that is actually being used<br /> <br /> - why is this hard<br /> - what does &quot;current&quot; actually mean?<br /> - when is stuff used recently not going to be used again?<br /> - past predicts the future, but not always<br /> <br /> - computer memory is designed assuming locality of various kinds<br /> - temporal locality<br /> - access in the past means likely access in the future<br /> - spatial locality (closer together data is more likely to be<br /> accessed at the same time)<br /> <br /> - code and data exhibit different locality patterns<br /> e.g. video streaming<br /> - run the same code over and over again<br /> - data is accessed once then discarded<br /> <br /> - to take advantage of locality patterns, we have a memory hierarchy<br /> - different types of memory, with different characteristics:<br /> - volatility<br /> - size<br /> - latency of access<br /> - bandwidth<br /> <br /> - most obvious: disk and RAM<br /> - disk is slow, but durable<br /> - RAM is fast but ephemeral<br /> <br /> - latency versus bandwidth<br /> - in general, you want low latency and high bandwidth<br /> - latency: time to get first byte of data<br /> - bandwidth: rate of data transfer over time<br /> - you can easily increase bandwidth by increasing parallelism<br /> - you have to design for latency<br /> <br /> - but real hierarchy is deeper<br /> - CPU registers<br /> - TLB, translation lookaside buffer<br /> - cache of virtual to physical memory mappings<br /> - it is an associative array implemented in hardware<br /> - L1 cache (smallest, lowest latency)<br /> - L2 cache (larger, moderate latency)<br /> - L3 cache (even larger, higher latency, but still faster than DRAM)<br /> - DRAM (pretty big, high latency, high bandwidth)<br /> - Solid state disks (flash memory)<br /> - Hard disks (spinning)<br /> - Tape<br /> <br /> - virtual versus physical addresses<br /> - programs access memory in terms of virtual addresses<br /> - storage happens with physical addresses<br /> - virtual to physical address translation happens<br /> ON EVERY MEMORY ACCESS<br /> &lt;/pre&gt;</div>

Soma