wow, lots of akamai

Thu Apr 1 20:56:56 UTC 2021

Tom,

All due respect, but there is a massive difference between one user
downloading 50G and thousands of users each downloading 50G when they all
go to play their videogame of choice at around the same time.

-Matt

On Thu, Apr 1, 2021 at 2:46 PM Tom Beecher <beecher at beecher.cc> wrote:

> A user sends a few megabytes of request and receives 50 gigs of reply.
>> They aren't DDoSing the network, but they're amplifying a single 50 gig
>> copy they receive from the mothership and turning it into likely tens of
>> terabytes of traffic.
>> Yes, that's a CDN's job, but that volume of legitimate traffic and the
>> very tiny window with which it is transmitted is likely to be a burden for
>> even the largest residential ISPs.
>>
>
> I'm sitting at home, and I could send a 50k request for a 50G file right
> now from a source not fronted by a CDN. What do? My ISP is still has to
> deliver it to me. The fact that the 50G file does or does not come from a
> CDN is irrelevant. The CDN just happens to be a point source that a lot of
> users happen to connect to.
>
> CDNs want to have the best performance to users because that's what brings
> them business. A poorly performing CDN will lose customers to a better
> performing one. The trend for years has been instead of ISPs investing in
> infrastructure to effectively handle the traffic that their users request,
> they turf that to CDNs. In many cases, a CDN will put a cache box in or
> extend a circuit at a loss to them, because they know if the performance
> metrics get bad, business will be taken elsewhere, even if the CAUSE of the
> poor performance is actually at the edge of, or inside , the ISPs network.
>
> ISPs in the US can get away with this because their users are captive and
> rarely have an alternative choice of provider.
>
>
> On Thu, Apr 1, 2021 at 4:33 PM Matt Erculiani <merculiani at gmail.com>
> wrote:
>
>> Patrick,
>>
>> > First, to be blunt, if you really think Akamai nodes are “sitting idle
>> for weeks” before CoD comes out with a new game,
>> > you are clearly confused.
>>
>> "Idle" in the sense that when you look at a graph of traffic before and
>> after a large push such as this makes the rest of the week's traffic look
>> like a horizontal line at the bottom, admittedly poor word choice, yes, but
>> far from "confused" as to what CDNs do under relatively normal
>> circumstances. Otherwise very valid points you've raised.
>>
>> Tom,
>>
>> > Akamai, and other CDNs, do not **generate** traffic ; they serve the
>> requests generated by users.
>>
>> A user sends a few megabytes of request and receives 50 gigs of reply.
>> They aren't DDoSing the network, but they're amplifying a single 50 gig
>> copy they receive from the mothership and turning it into likely tens of
>> terabytes of traffic.
>> Yes, that's a CDN's job, but that volume of legitimate traffic and the
>> very tiny window with which it is transmitted is likely to be a burden for
>> even the largest residential ISPs.
>>
>> -Matt
>>
>> On Thu, Apr 1, 2021 at 2:09 PM Patrick W. Gilmore <patrick at ianai.net>
>> wrote:
>>
>>> Matt:
>>>
>>> I am going to disagree with your characterization of how Akamai - and
>>> many other CDNs - manage things. First, to be blunt, if you really think
>>> Akamai nodes are “sitting idle for weeks” before CoD comes out with a new
>>> game, you are clearly confused.
>>>
>>> More importantly, I know for a fact Akamai has spent ungodly amounts of
>>> money & resources putting content precisely where the ISPs ask them to put
>>> it, deliver it over the pipes the ISPs ask them to deliver it, at precisely
>>> the capacity the ISPs tell them.
>>>
>>> On the other hand, I agree with your characterization of residential
>>> broadband. It is ridiculous to expect a neighborhood with 1,000 homes each
>>> with 1 Gbps links to have a terabit of uplink capacity. But it also should
>>> have a lot more than 10 Gbps, IMHO. Unfortunately, most neighborhoods I
>>> have seen are closer to the latter than the former.
>>>
>>> Finally, this could quickly devolve into finger pointing. You say the
>>> CDNs bear some responsibility? They may well respond that the large
>>> broadband providers ask for cash to interconnect - but still require the
>>> CDNs to do all the work. The CDNs did not create the content, or tell the
>>> users which content to pull. When I pay $NATIONAL_PROVIDER, I expect them
>>> to provide me with access to the Internet. Not just to the content that
>>> pays that provider.
>>>
>>> Personally, I have zero problems with the ISPs saying “give me a cache
>>> to put here with this sized uplink” or “please deliver to these users over
>>> this xconn / IX / whatever”. I have a huge problem with the ISPs blaming
>>> the ISPs for delivering what the ISP’s users request.
>>>
>>> Of course, this could all be solved if there were more competition in
>>> broadband in the US (and many other countries). But that is a totally
>>> different 10,000 post thread (that we have had many dozens of times).
>>>
>>> --
>>> TTFN,
>>> patrick
>>>
>>> On Apr 1, 2021, at 3:53 PM, Matt Erculiani <merculiani at gmail.com> wrote:
>>>
>>> Niels,
>>>
>>> I think to clarify Jean's point, when you buy a 300mbps circuit, you're
>>> paying for 300mbps of *internet *access.
>>>
>>> That does not mean that a network should (and in this case small-medium
>>> ones simply can't) build all of their capacity to service a large number of
>>> customer circuits at line rate at the same time for an extended
>>> period, ESPECIALLY to the exact same endpoint. It's just not economically
>>> reasonable to expect that. Remember we're talking about residential service
>>> here, not enterprise circuits.
>>>
>>> Therefore, how do you prevent this spike of [insert large number here]
>>> gigabits traversing the network at the same time from causing issues? Build
>>> more network? That sounds easy, but there are plenty of legitimate reasons
>>> why ISPs can't or don't want to do that, particularly for an event that
>>> only occurs once per quarter or so.
>>>
>>> Does Akamai bear some burden here to make these rollouts less
>>> troublesome for the ISPs they traverse through the last mile(s)? IMO yes,
>>> yes they do. When you're doing something new and unprecedented, as Akamai
>>> frequently brags about on Twitter, like having rapid, bursty growth of
>>> traffic, you need to consider that just because you can generate it,
>>> doesn't mean it can be delivered.  They've gotta be more sophisticated than
>>> a bunch of servers with SSD arrays, ramdisks, and 100 gig interfaces, so
>>> there's no excuse for them here to just blindly fill every link they have
>>> after sitting idle for weeks/months at a time and expect everything to come
>>> out alright and nobody to complain about it.
>>>
>>> On Thu, Apr 1, 2021 at 1:21 PM Niels Bakker <niels=nanog at bakker.net>
>>> wrote:
>>>
>>>> * nanog at nanog.org (Jean St-Laurent via NANOG) [Thu 01 Apr 2021, 21:03
>>>> CEST]:
>>>> >An artificial roll out penalty somehow? Probably not at the ISP
>>>> >level, but more at the game level. Well, ISP could also have some
>>>> >mechanisms to reduce the impact or even Akamai could force a
>>>> >progressive roll out.
>>>>
>>>> It's an online game. You can't play the game with outdated assets.
>>>> You'd not see walls where other players would, for example.
>>>>
>>>> What you're suggesting is the ability of ISPs to market Internet access
>>>> at a certain speed but not have to deliver it based on conditions they
>>>> create.
>>>>
>>>>
>>>>         -- Niels.
>>>>
>>>
>>>
>>> --
>>> Matt Erculiani
>>> ERCUL-ARIN
>>>
>>>
>>>
>>
>> --
>> Matt Erculiani
>> ERCUL-ARIN
>>
>

-- 
Matt Erculiani
ERCUL-ARIN
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.nanog.org/pipermail/nanog/attachments/20210401/44bdbfbc/attachment.html>