Episode Details

Papers we love: ARC by Bryan Cantrill, SSD caching adventures with ZFS, OpenBSD full disk encryption setup, and a Perl5 Slack Syslog BSD daemon.

This episode was brought to you by

Headlines

Papers We Love: ARC: A Self-Tuning, Low Overhead Replacement Cache

• Ever wondered how the ZFS ARC (Adaptive Replacement Cache) works?
• How about if Bryan Cantrill presented the original paper on its design?
• Today is that day.
• Slides
• It starts by looking back at a fundamental paper from the 40s where the architecture of general-purpose computers are first laid out
• The main is the description of memory hierarchies, where you have a small amount of very fast memory, then the next level is slower but larger, and on and on. As we look at the various L1, L2, and L3 caches on a CPU, then RAM, then flash, then spinning disks, this still holds true today.
• The paper then does a survey of the existing caching policies and tries to explain the issues with each. This includes ‘MIN’, which is the theoretically optimal policy, which requires future knowledge, but is useful for setting the upper bound, what is the best we could possibly do.
• The paper ends up showing that the ARC can end up being better than manually trying to pick the best number for the workload, because it adapts as the workload changes
• At about 1:25 into the video, Bryan start talking about the practical implementation of the ARC in ZFS, and some challenges they have run into recently at Joyent.
• A great discussion about some of the problems when ZFS needs to shrink the ARC. Not all of it applies 1:1 to FreeBSD because the kernel and the kmem implementation are different in a number of ways
• There were some interesting questions asked at the end as well ***

How do I use man pages to learn how to use commands?

• nwildner on StackExchange has a very thorough answer to the question how to interpret man pages to understand complicated commands (xargs in this case, but not specifically).
• Have in mind what you want to do.

When doing your research about xargs you did it for a purpose, right? You had a specific need that was reading standard output and executing commands based on that output.

• But, when I don't know which command I want?

Use man -k or apropos (they are equivalent). If I don't know how to find a file: man -k file | grep search. Read the descriptions and find one that will better fit your needs.

• Apropos works with regular expressions by default, (man apropos, read the description and find out what -r does), and on this example I'm looking for every manpage where the description starts with "report".

• Always read the DESCRIPTION before starting

• Take a time and read the description. By just reading the description of the xargs command we will learn that:

xargs reads from STDIN and executes the command needed. This also means that you will need to have some knowledge of how standard input works, and how to manipulate it through pipes to chain commands