All things related to multicast

[Alan Crosswell <>]

Yeah, back in the dot-com bubble days, Akamai installed 14 or 15 servers
here which seemed like pretty massive overkill.  A few years ago they
removed 10 or 11 of them and presumably redeployed them elsewhere.  So I
suspect we don't have the full suite.  I suppose we could span the
subnet and see what's flowing....  I don't think we've ever seen better
than a 4 or 5 to 1 ratio on traffic out to users vs. in from upstream.
/a

Hank Nussbacher wrote:
> On Sat, 24 Jan 2009, Alan Crosswell wrote:
>
> Hi Alan!  Been what - 27 years?  In any event, I once heard from
> Akamai that they have "different" stacks deployed - at academia they
> have servers that serve up static content whereas at ISPs they have
> servers that serve up streaming and dynamic content.  We too did not
> see any jump in Akamai traffic at our 3 server stack:
> http://noc.ilan.net.il/stats/TEL-AVIV-GP/akamai_hp_switch.html
>
> -Hank
>
>> Most of the popular news sights like CNN are not using Akamai.  We saw
>> only a minor increase in bandwidth coming out of our Akamai stack on
>> campus of about 150 Mbps (vs. about 40 Mbps into it).  This was only
>> maybe a 10-20% increase over a typical day.  Now maybe our on-campus
>> stack is only doing static pages and not stream caching.  It's hard to
>> tell what those boxes are up to.
>> /a
>>
>> Hank Nussbacher wrote:
>>> On Fri, 23 Jan 2009, Julian Y. Koh wrote:
>>>
>>> Obama inauguration sets Web traffic record, Akamai says
>> http://www.networkworld.com/news/2009/012109-obama-inauguration-web-traffic.html
>>
>>>
>>> -Hank
>>>
>>> OK, some quick and dirty stats: on any given day that's not the
>>> historic
>>> inauguration of a new US President, we have 1-2 people watching
>>> C-SPAN or
>>> C-SPAN2.
>>>
>>> On 1/16, we peaked up at about 150 simultaneous clients logged in as
>>> people
>>> tested their multicast connectivity and client compatibility. 
>>> Throughout
>>> the weekend and Monday we had about 30-40 clients connected at any
>>> given
>>> time.
>>>
>>> Things started taking of on Tuesday by 7am Central, when we had ~100
>>> clients connected, and the rate of growth was rather steady up to ~1000
>>> clients right before 10am.  Then the rate of grown increased again
>>> at was
>>> steady up to a little over 2000 clients right before 11am.  That was
>>> our
>>> peak, where we stayed until 11:30am, which I believe is when Obama
>>> stopped
>>> speaking.  Then we have a very quick drop back down to ~1000 clients by
>>> 12:20pm, then a slowly decreasing count down to ~500 by 3pm, then
>>> leveling
>>> off back at ~100 clients by 8pm.
>>>
>>> Throughout it all, the encoder/license server never broke a sweat,
>>> staying
>>> at ~.1 load average (the encoding process is mostly
>>> hardware-accelerated,
>>> but all the database/stats/licensing is CPU-driven).  Obviously our
>>> outbound bandwidth related to this also stayed at a steady ~5.3Mbps
>>> or so
>>> (C-SPAN1 is 2.1Mbps, C-SPAN2 is ~3.2Mbps).
>>>
>>> Now, we had definite pains on some of our external traffic due to the
>>> Octoshape P2P streaming that CNN was employing, but luckily we
>>> already had
>>> our multicast traffic bypassing the equipment that was hurting. 
>>> Otherwise
>>> life would have been not so good for us.  As it was, the pain that
>>> people
>>> were experiencing with the unicast streams (and I'm including the
>>> Octoshape
>>> streams there as well) is going to serve as a good leverage for
>>> publicizing
>>> the multicast-based services.
>>>
>>>
>>>>
>> -- 
>> Julian Y. Koh                         
>> <mailto:>
>> Network Engineer                                   <phone:847-467-5780>
>> Telecommunications and Network Services         Northwestern University
>> PGP Public Key:<http://bt.ittns.northwestern.edu/julian/pgppubkey.html>
>>>>
>>
>>
>