About
I use this blog to post things I find most interesting and important, and very occasionally to unleash a rant.
As to who I am, why trust me when you can trust others?
As a quick synopsis, Bob Scheifler and I started the X Window System; I was the editor of the HTTP/1.1 proposed and draft standards, co-founder of handhelds.org and the revised X.org, and helped out on freedesktop.org. I also worried a lot about the software for the OLPC XO-1. And a long time ago, in a galaxy far, far away, I used to astronomy.
jg's Ramblings 
December 7, 2010 at 9:49 pm |
Hi Jim,
What was the MTU in your buffer bloat tests? Also did you try varying the MTU to see if it made a difference in the buffering capabilities?
Jim
December 7, 2010 at 10:06 pm |
No, I didn’t try playing with the MTU. I expect it was blog standard out of the box. Go look at my packet captures.
December 15, 2010 at 8:51 am |
Since you’re probably going to write about AQMs in a future installment, please let me point you at SFB,
http://www.pps.jussieu.fr/~jch/software/sfb/
A rather nifty AQM that I’ve never managed to get into the kernel tree.
–jch
December 15, 2010 at 9:07 am |
Yes, I will be writing about AQM’s shortly. The major issue with AQM’s has been needing tuning; ergo they haven’t been trusted, and aren’t always enabled. I’ll elaborate some when I post about that.
Good to hear from you Juliusz, in any case.
January 3, 2011 at 4:39 pm |
Hi Juliusz,
Interesting qdisc. What have kept you back from getting this stuff into mainline?
I have not mentioned this before, but I work for a Danish ISP (www.comx.dk) where I have implemented a Linux based bandwidth shaper. Our customers does not experience the buffer bloat problem, because I have given each customer their own SFQ (Stochastic Fairness Qdisc) which tries to give each flow a fair share.
Perhaps your SFB (Stochastic Fair Blue) implementation is a better choice! 🙂
January 5, 2011 at 5:15 pm |
> What have kept you back from getting this stuff into mainline?
Lack of energy. After spending two weeks trying to explain why the existing packet-hashing infrastructure in the kernel was not adapted to being used with Bloom filters to people who were clearly not interested in listening, I decided that I had better things to do with my life.
http://lwn.net/Articles/277208/
http://lwn.net/Articles/277802/
> I have given each customer their own SFQ
Note that SFQ and SFB do very different things. SFB never reorders packets — it only “stochastically fairly” drops/marks packets, while SFQ reorders packets, but never drops them (unless the parent Qdisc drops them, of course).
Hence SFQ is good at ensuring fairness, but doesn’t do anything towards limiting buffer bloat; on the other hand, SFB is not particularly good at being fair (it’s resource fair, not throughput fair), but it’s brilliant at limiting buffer sizes. Which is what I am interested in, as I work on mesh networks.
–jch
January 8, 2011 at 2:09 am |
What is old is new again. See John Nagel’s RFC-970.
January 8, 2011 at 10:06 am |
Hi Mike.
Yup. John Nagle was right on. And we now have a network with (approaching) infinite storage. We shouldn’t be surprised at the result…
New (to some of us, probably not to you and Van) is the idea that the buffers destroy congestion avoidance as well.
Thanks for reminding me of this document; I’ll add it to the FAQ I’ve been building, as its exposition is really quite nice.
March 28, 2012 at 11:40 pm |
Hi jim, im a student of computer science in Chile, and as my thesis work, i want to test the presence and effects of bufferbloat in different configurations. I wonder if i can share my doubts, questions and doubts with you and with the community. i main idea is to replicate your tests and verify the existence of this problem. Also test the cerowrt firmware.
Thanks for share and all the well explained info.
Juan S. Catalan
May 2, 2012 at 8:07 pm |
Juan: Did you join the bloat list? be glad to have more thesis students around…
May 2, 2012 at 9:37 pm |
I did, and i try to keep it up with all the news and things related!… Once i got some results of the test that i’m planning to do, i will be glad to share them with all the community.
Juan S. Catalan
May 2, 2012 at 9:42 pm |
Also Dave, if you want to practice your spanish, es cosa que me envies un correo 😉
saludos!
Juan S. Catalán