Ethernet polling and patch-pulling latency [LWN.net]

Ethernet polling and patch-pulling latency

Posted Jul 11, 2013 9:58 UTC (Thu) by niner (guest, #26151)
Parent article: Ethernet polling and patch-pulling latency

Is POLL_BUSY_LOOP really a better name? Sure it does spell out the drawbacks very clearly, but it does so naming an implementation detail. Maybe at some point in the future we find another way to provide low latency device polling not using busy looping. Maybe it will be some fancy hardware feature or some new genius trick with new drawbacks. Applications probably will not care about how low latency is achieved, but the interface will then be misleading.

to post comments

Ethernet polling and patch-pulling latency

Posted Jul 11, 2013 18:42 UTC (Thu) by error27 (subscriber, #8346) [Link]

Naming is hard, but if those are the only two choices then POLL_BUSY_LOOP is better. The other name is too tempting and you'll always be having to explain it and why no one outside the financial industry should use it. By naming it something scary you've cut your support costs dramatically.

Honestly, there will always be people who want to busy loop, so you're right that it's an implementation detail but it's also probably not going to change.

Ethernet polling and patch-pulling latency

Posted Jul 11, 2013 21:34 UTC (Thu) by iabervon (subscriber, #722) [Link]

If there aren't drawbacks, the networking stack would just start using the feature regardless of what the application specifically cares about. If there are drawbacks, it might matter what they are. I mean, you can often improve latency by delivering corrupted packets instead of waiting for correct copies, but that's not a tradeoff everyone who wants low latency is willing to make.