[Codel] Some small bits re the lwn draft article review

David Collier-Brown davec-b at rogers.com
Wed Dec 5 09:18:21 EST 2012

After flushing what I know out of my head, I had a look at the article,
and noticed a few clarifications you might want to consider for the LWN

In "SFQ Overview", you dive into mechanism, but you might leave the new
reader behind.  Instead of "With high probability, isolate “hog”
sessions so that they bear the brunt of any packet dropping that might
be required. To this end, ...", how about something more like "With high
probability, isolate individual sessions into individual queues, ten
arrange that the ones with the "hogs" bear the brunt of any packet
dropping that might be required.  To implement this,"...

I also wonder if one would want to use "flow" instead of "session"? It
does have some specific technical senses, but the general sense seems to
fit well here.

Further down in the same section, you note "There is also an index into
this array that tracks the buckets with the most packets."  I'd suggest
you tie back to the motivation, and add "From this we can easily
distinguish the hogs from the low-bandwidth flows." (or sessions).

A but further (in para 8), you mention that SFQ is "often" configured in
Linux. It might be better to say it appeared in some specific version
and/or distribution at a particular date...

In "FQ-CoDel Overview", para 4, you say "Nevertheless, CoDel still
maintains a single queue". Since you're about to distinguish FQ-CoDel
from it, I'd suggest emphasizing the variant, with "Nevertheless, the
original CoDel only maintains a single queue"

In "FQ-CoDel Overview", para 5, you say "packets are hashed into
multiple buckets", which could be misconstrued. Perhaps it could be
"packets are hashed into individual buckets"

Much further on, after QQ 5, you say "This is extremely important: if
there was an unending drizzle of random packets, eventually...", I
simply don't get it. Could you expand a bit more here?

Just before QQ 7, you write "FQ-CoDel gives users a choice between low
latency and high reliability on ..".  I'd suggest the user's don't get
the choice, but rather the packets do, so a better phrasing might be "
FQ-CoDel deals differently with low latency and high reliability flows
(or sessions) on the one hand ..."

In "Effectiveness of FQ-CoDel", in para 2, you define BK CS1/5 and EF in
terns of what they stand for. Regrettably, the names don't say what the
selectors are supposed to achieve, so the reader doesn't know which one
is supposed to be the best for low latency versus high throughput, and
will therefor fail to see part of the benefit of FQ_CoDel.

In "Remaining Challenges", In the WiFi paragraph, you say "addressing
bufferbloat in the presence of WiFi", which suggests that having WiFi on
one interface or part of ones network will degrade others. Perhaps "over
WiFi" would be more appropriate?

--dave c-b
David Collier-Brown,         | Always do right. This will gratify
System Programmer and Author | some people and astonish the rest
davecb at spamcop.net           |                      -- Mark Twain
(416) 223-8968

More information about the Codel mailing list