[142488] in North American Network Operators' Group
Re: Announcing Project BISMark: ISP Performance Measurements from
daemon@ATHENA.MIT.EDU (Nick Feamster)
Tue Jun 28 10:58:47 2011
From: Nick Feamster <feamster@cc.gatech.edu>
In-Reply-To: <alpine.DEB.2.00.1106281209420.19581@uplift.swm.pp.se>
Date: Tue, 28 Jun 2011 10:58:32 -0400
To: Mikael Abrahamsson <swmike@swm.pp.se>
Cc: nanog@nanog.org
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
Thanks for the feedback!
On Jun 28, 2011, at 6:13 AM, Mikael Abrahamsson wrote:
> On Mon, 27 Jun 2011, Nick Feamster wrote:
>=20
>> We've launched Project BISMark, a project that performs active =
performance measurements of upload and download throughput, latency, =
etc. from OpenWRT-based routers running inside of homes. We have tested =
our OpenWRT image on the NetGear WNDR 3700v2 and are currently shipping =
out NetGear routers with the BISMark firmware to anyone who is =
interested.
>=20
> Please, pretty please, with sugar on top, don't just do active =
measurement, but also do passive measurement of real traffic. Doing test =
traffic is one case, but the really important thing to look at is real =
traffic. I tried to get traction for this on IETF75, but there seems to =
be little interest.
>=20
We would very much like to. There are a number of reasons that regular =
users seem to be asking for passive measurement, such as monitoring of =
traffic usage of different applications (e.g., "How much is streaming =
eating into my usage cap?"). A few years ago, we had a tool that would =
do all of this with passive measurement (http://gtnoise.net/nano/), and =
we'd certainly like to resume this line of inquiry, if we can figure out =
how to address people's privacy concerns. We are developing a passive =
measurement suite for BISMark, which is also available on github.
> On a NAT router there is a state table, what would the performance =
penality be to look at TCP sequence numbers, RTTs (TCP timestamps) to be =
able to discern PDV and loss of the actual traffic the customer is =
doing?
>=20
> There are a lot of test suites, they solve one problem, but a passive =
monitoring system that would show how the real traffic is behaving would =
yield a lot more valuable information that just relying on active =
testing (which will cause harm to customer traffic when the test is =
run).
Definitely good points, and we've thought about this, for sure. The key =
question seems to be how to handle user privacy in a way that everyone =
can be happy with. Ultimately, we might take a survey of our users =
about this (e.g., certain people have said they don't mind tracking =
application performance/usage as long as the specific Web sites or =
destinations are not logged). It would be really helpful to get an =
understanding of what users might find acceptable, as far as passive =
measurements.
-Nick=
