[Bloat] viability of the data center in the internet of the future

David P. Reed dpreed at reed.com
Sat Jun 28 08:31:51 EDT 2014


I hope it is obvious I am in violent agreement. 

The folks who think a centralized structure is more efficient or more practical just have not thought it through.

The opposite is true. Sadly people's intuitions are trained to ignore evidence and sound argument....

So we have a huge population of engineers who go along without thinking because they honestly think centralized systems are better for some important reason they never question.

This means that the non engineering public has no chance at understanding.

Whenever I have looked at why centralized designs are 'needed' it has turned out to be the felt need for 'control' of something by one small group or individual.

Sometimes it is the designer. Shame on him/her. 

Sometimes it is the builder. Ditto.

Sometimes it is the operator. Do we need one operator?

Sometimes it is the owner. Don't the users own their uses and purposes?

Sometimes it is the fearful. I sympathize. But they don't really want to cede collective control to a small group they can't trust ir even understand. Or maybe they do...

Sometimes it is the wannabe sovereign. 

The weakness of the argument is that control need not be centralized. In fact centralized control is inefficient and unnecessary.

I've devoted much of my work to that last sentence.

For example... In Croquet we (me and 3 others) demonstrated that it's pretty easy to build a real time shared multimedia virtual world that works without a single central server. It really worked and scaled linearly with users adding their own computer when they entered the world, and removing it when they got disconnected. . Just pulling the plug was fine.) 

Same with decentralized wireless ... no need for centralized spectrum allocation... linear growth of capacity with transceiver participation coming from the actual physics of the real propagation environment.

Equating centralized control with efficiency or necessary management is a false intuition.

Always be skeptical of the claim that centralized control is good. Cui bono?

On Jun 28, 2014, Dave Taht <dave.taht at gmail.com> wrote:
>I didn't care for my name in the subject line in the first place,
>although it did inspire me to do some creative venting elsewhere, and
>now here. And this is still way off topic for the bloat list...
>
>One of the points in the wired article that kicked this thread off was
>this picture of what the internet is starting to look like:
>
>http://www.wired.com/wp-content/uploads/2014/06/net_neutral.jpg.jpeg
>
>I don't want it to look like that. I worked pretty hard to defuse the
>"fast vs slow" lane debate re peering because it was so inaccurate,
>and it does look like it has died down somewhat, but
>that doesn't mean I like the concentration of services that is going
>on.
>
>I want the "backbone" to extend all the way to the edge.
>
>I want the edge to be all connected together, so in the unlikely event
>comcast goes out of business tomorrow, I can get re-routed 1 hop out
>from my house through verizon, or joe's mom and pop fiber shop, or
>wherever. I want a network that can survive multiple backhoe events,
>katrinas, and nuclear wars, all at the same time. I'd like to be able
>to get my own email,
>and do my own phone and videoconferencing calls with nobody in the
>middle, not even for call setup, and be able to host my own my own
>services on my own hardware, with some level of hope that anything
>secret or proprietary stays within my premise. I want a static ip
>address range, and
>control over my own dns.
>
>I don't mind at all sharing some storage for the inevitable
>advertising if the cdn's co-located inside my business are also
>caching useful bits of javascript, etc, just so I can save on latency
>on wiping the resulting cruft from my eyeballs. I want useful
>applications, running, directly, on my own devices, with a minimum
>amount of connectivity to the outside world required to run them. I
>want the 83 items in my netflix queue already downloaded, overnight,
>so I can pick and choose what to see without ever having a "Buffering"
>event. I want my own copy of wikipedia, and a search engine that
>doesn't share everything you are looking for with the universe.
>
>I want the legal protections, well established for things inside your
>home, that are clearly not established in data centers.
>
>I'd like it if the software we had was robust, reliable, and secure
>enough to do that. I'd like it if it were easy to make offsite
>backups, as well as mirror services with friends and co-authors.
>
>And I'd like my servers to run on a couple watts, at most, and not
>require special heating, or cooling.
>
>And I'd like (another) beer and some popcorn. Tonight's movie:
>
>https://plus.google.com/u/0/107942175615993706558/posts/VJKvfvKU9pi
>
>On Fri, Jun 27, 2014 at 9:28 PM, Dave Taht <dave.taht at gmail.com> wrote:
>> On Fri, Jun 27, 2014 at 9:06 PM, David P. Reed <dpreed at reed.com>
>wrote:
>>> Maybe I am misunderstanding something... it just took my Mac book
>Pro doing
>>> an rsync to copy a TB of data from a small NAS  at work yesterday to
>get
>>> about 700 Gb/sec on a GigE office network for hours yesterday.
>>>
>>> I had to do that in our Santana Clara office rather than from home
>outside
>>> Boston, which is where I work 90% of the time.
>>>
>>> That's one little computer and one user...
>>
>> On a daily basis, the bufferbloat websites transfer far, far less
>than gigE
>>
>> IF the redmine portion of the site wasn't so cpu expensive, I could
>> use something
>> other than hefty boxes they are on. Similarly snapon's cpu is mostly
>> used for builds, the file transfer role could be done by something
>else
>> easily. I'd like to switch it over to do that one day.
>>
>>> What does my Mac Book Pro draw doing that? 80 Watts?
>>
>> I love the "kill-a-watt" products. I use them everywhere. (while I'm
>> pimping stuff I like, digilogger's power switches are a lifesaver
>also -
>> staging boots for devices that draw a lot of power in a tiny lab that
>> can only draw 350 watts before becoming a fire hazard)
>>
>> Your NAS probably ate less than 16 watts, more if you have more than
>one drive.
>>
>> My nucs draw 18 watts and can transfer at GigE off a flash disk
>> without raising a sweat.
>> (at least some of your overhead is in the rsync protocol, which is
>> overly chatty)
>>
>> Several tiny arm boards can all do gigE at line rate, notably stuff
>built around
>> marvell and cavium's chipset(s), and they do it at under 2 watts.
>Most support
>> 64GB mini-sd cards (with pretty lousy transfer rates).
>>
>> Pretty sure (haven't booted it yet) the parallella (which is smaller
>> than a drive),
>> can do it in under a 2 watt, and if it doesn't do gigE now, it'll do
>> it after I get through
>> with it - but it lacks a sata port, and usb is only 2.0, so it might
>> not drive gigE
>> from a nas perspective. (It kind of bugs me that most of the tiny
>boards are in
>> the altoids form factor, rather than the 2.5 inch drive form factor)
>>
>> So I go back to my original point in that, once you have fiber to the
>business,
>> for most purposes in a small business or startup or home - who needs
>> to co-lo in a data center?
>> You can have a tiny wart on the wall do most of the job. And that's
>> today. In another
>> year or so we'll be over some more tipping points.
>>
>> One thing that does bug me is most UPSes are optimized to deliver a
>large
>> load over a short time, a UPS capable of driving 5 watts for, say, 3
>days is
>> kind of rare.
>>
>>> On Jun 27, 2014, David Lang wrote:
>>>>
>>>> On Tue, 24 Jun 2014, Michael Richardson wrote:
>>>>
>>>>> Rick Jones wrote:
>>>>>>
>>>>>> Perhaps, but where does having gigabit fibre to a business imply
>the
>>>>>> business
>>>>>> has the space, power, and cooling to host all the servers it
>might
>>>>>> need/wish
>>>>>> to have?
>>>>>
>>>>>
>>>>> That's a secondary decision.
>>>>> Given roof space, solar panels and/or snow-outside, maybe the
>answer is
>>>>> that
>>>>> I regularly have 2 our of 3 of those available in a decentralized
>way.
>>>>
>>>>
>>>> given the amount of processing capacity that you can get today in a
>>>> pasively
>>>> cooled system, you can do quite a b it of serving from a small
>amount of
>>>> space
>>>> and power.
>>>>
>>>> The days when it took rooms of Sun boxes to saturate a Gb line are
>long
>>>> gone,
>>>> you can do that with just a handful of machines.
>>>>
>>>> David Lang
>>>> ________________________________
>>>>
>>>> Cerowrt-devel mailing list
>>>> Cerowrt-devel at lists.bufferbloat.net
>>>> https://lists.bufferbloat.net/listinfo/cerowrt-devel
>>>
>>>
>>> -- Sent from my Android device with K-@ Mail. Please excuse my
>brevity.
>>>
>>> _______________________________________________
>>> Bloat mailing list
>>> Bloat at lists.bufferbloat.net
>>> https://lists.bufferbloat.net/listinfo/bloat
>>>
>>
>>
>>
>> --
>> Dave Täht
>>
>> NSFW:
>https://w2.eff.org/Censorship/Internet_censorship_bills/russell_0296_indecent.article

-- Sent from my Android device with K-@ Mail. Please excuse my brevity.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/bloat/attachments/20140628/d5cee046/attachment-0003.html>


More information about the Bloat mailing list