[Bloat] Bufferbloat in switches and routers

[Ketan Kulkarni]

On Apr 18, 2015 13:33, "David Lang" <david at lang.hm> wrote:
>
> On Sat, 18 Apr 2015, Ketan Kulkarni wrote:
>
>> Hi,
>> We have been talking about the bloated buffers mostly on the home
routers.
>> The Cisco PIE too has been standardized by docsis meant to be for cable
>> modems
>>
>> I think we would have similar concerns for switches and routers. (E.g.
>> cat3k switches or Cisco 5760 controllers just to name)
>
>
> remember that bufferbloat shows up where there is a difference in
bandwidth from one side of the router to the other (i.e. a bottleneck)
>
Thanks this makes the devices easier to target.

> This is almost always going to happen at the edge of your LAN where you
go from your Gig-E (or in a datacenter, possibly 10Gig-E to your WAN link.
It can happen at places inside your datacenter, but isn't as likely
>
Agree. As per my (limited) knowledge of such deployment goes, these
probably never run to their peak capacity in the production/live system
probably not even to saturate the lowest of the links( I may be wrong
though) . Given this, what is the gravity of the effect of the bufferbloat?
Or such study has never been done before?

Having AQM won't definitely hurt, however is it indeed a real problem to
solve for such edge routers?

At the same time I hear codel or pie finding the space in data centers. So
there are definitely some pieces I am missing.

Sorry for being little naive but the answers will definitely help me
understand the problem space and spread more awareness about bufferbloat.

>
>> I would like to know your views about what you think about it .
>> Are the theories so far and the AQMs  (codel and pie) stand true for such
>> devices too?
>
>
> If they are bottlenecks, yes. If they are not bottlenecks it won't hurt
(no queues will build up
>
>
>> What would it take to measure the bloat levels of these devices? Do we
>> still need to use the netperf wrapper to get the characteristics of such
>> devices?
>
>
> the same approach works. you may need beefier systems to generate
sufficient load.
>
> David Lang
> _______________________________________________
> Bloat mailing list
> Bloat at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/bloat
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/bloat/attachments/20150418/0ade8f37/attachment-0002.html>