[Make-wifi-fast] I used to dream of a single wifi cpu, memory, and I/O

[David Lang]

you can't just treat wifi as any other transport and have things work well. 
There is too much overhead in sending packets to any destination. Every other 
transport has effectively zero cost to send a packet to A a packet to B and then 
another packet to A. In wifi, you can send hundreds (if not thousands) of 
packets to A in the time that it will take you to send a packet to B. As a 
result, you need to queue things per destination in ways that you don't for 
anything else.

To see how this works, look at the time that you wait to see if the channel is 
used, plus the time needed to transmit the packet header (at a low bitrate), and 
then the time needed to send the actual data (at a high bitrate)

see how much actual data you could send to one station in the time that it would 
take to send two 64 byte packets to two different stations (ignoring MU-MIMO for 
the moment)

Also, as a practical matter, under most conditions, almost all your traffic is 
going to be wifi <-> wired, very little is wifi <-> wifi (IoT could have changed 
this, but in practice, IoT is IoT <-> cloud <-> user rather than IoT <-> user)

David Lang


On Mon, 10 Jul 2023, Bob McMahon via Make-wifi-fast wrote:

> Date: Mon, 10 Jul 2023 13:47:32 -0700
> From: Bob McMahon via Make-wifi-fast <make-wifi-fast at lists.bufferbloat.net>
> Reply-To: Bob McMahon <bob.mcmahon at broadcom.com>
> To: Aaron Wood <woody77 at gmail.com>
> Cc: Make-Wifi-fast <make-wifi-fast at lists.bufferbloat.net>
> Subject: Re: [Make-wifi-fast] I used to dream of a single wifi cpu, memory,
>     and I/O
> 
> I think of it as a micro translational bridge, i.e. get 802.11 to 802.3 to
> standard forwarding silicon (though some call it a "skinny" AP.) The "AP
> split" is between managing MAC/frame things and MAC/PHY things. Try to
> minimize the PHY knobs so the RRH can self-manage what's not been preset by
> the installer. Remove the sw config stuff on the RRH too as it just adds
> OPEX costs for limited benefit.
>
> Others see it as "bluetooth p2p" model applied to WiFi,  but radio pairings
> at low latency, low power and high throughputs - though this mental model
> has some flaws too.
>
> I think the EAP architecture is woefully broken. "Centralizing the
> controllers" of distribution management isn't the same as *eliminating &
> mitigating the need to manage complex distributed things.  *Linux computers
> are complex, distributed things and don't belong in homes unless the home
> is owned by a network and sys admin ;)
>
> The closest models seem to be ECPRI used by 5G towers or the distributed
> access architecture (DAA) of HFC.
>
> FiWi will blow those out of the water by my judgment - though those are OSP
> things and FiWi will start as an inside plant thing.  We just have to
> figure out how to motivate humans as engineers & marketing experts to start
> moving in this direction. It seems like a no brainer that consumers will
> want to show off their "fiber" - inside plant, future proof, life support
> capable, resilient to failures, & something that increases the property or
> asset value of their primary investment. Maybe FTTH shouldn't be marketed
> as "Fiber to the Home" but "Fiber through the Home" with no more wires for
> comm anywhere?
>
> Bob
>
>
>
>
> On Mon, Jul 10, 2023 at 12:02 PM Aaron Wood <woody77 at gmail.com> wrote:
>
>>
>>
>> On Sun, Jul 9, 2023 at 10:40 PM Bob McMahon via Make-wifi-fast <
>> make-wifi-fast at lists.bufferbloat.net> wrote:
>>
>>> I wonder if treating WiFi like a transceiver is the better approach,
>>> separating "the computer abstraction" from the remote radio head.
>>>
>>> Then it's a RF Front End(s) / CMOS Radios / PHY(s) / 802.11 MAC(s)
>>> (lower) / 802.3 PAM4 / 100G SERDES / 4 1x25G VCSELs. Pluggable like an SFP.
>>>
>>> The virtualized APs could support at least 10 and maybe 100s of these via
>>> merchant switching silicon and could be kilometers away. The MACs lower
>>> could be simplified to dual queue a la L4S. No need for 4 AC queues per
>>> MAC. Might be able to throw away 802.11 retries too and let the upper
>>> layers handle it.
>>>
>>
>> Isn't this how the commercial APs that use a combined
>> backhaul/control-plane operate?
>>
>> I've often wanted to use something like OpenVSwitch to combine all the
>> layer-2 broadcast domains without actually making them a single broadcast
>> domain (central ARP responder/router, multicast-to-singlecast conversion,
>> mDNS routing instead of always using broadcasts, etc).  I know that we were
>> looking at that with CeroWRT, early on, using routing between separate
>> subnets, but that required a bunch of proxies that never seemed to really
>> work as well as I wanted (or I wasn't very good at getting them set up
>> correctly).  I think we're maybe in a better place for that now.
>>
>> Given that wifi stations must associate with APs, there's so much
>> knowledgeable state that can be centrally (and distributedly) cached and
>> then used to move a bunch of broadcast traffic to single-cast.  With the
>> huge difference in encoding rates between broadcast and singlecast traffic,
>> especially as the number of STAs increases, it seems like it would be very
>> beneficial.
>>
>> And if you were in the 1:1 AP<->STA situation that has floated across this
>> list a few times (I do find the micro-AP idea fascinating), then all
>> "broadcast" traffic should hopefully become limited-transmit-power
>> single-cast, with minimal use of spectrum (both in terms of time and
>> physical space).
>>
>> One issue that I've seen with many APs in the same space, is that many
>> client stations are very aggressive enumerators of the available APs, much
>> to the detriment of the operation of the system.  I've seen clients DoS all
>> the available spectrum with wildcard probe requests and their responses,
>> which sound like exactly the sort problems seen with ARP in very dense
>> server racks that the vswitch and central ARP server model was meant to
>> solve:  have a single "best" response instead of flooding the network with
>> responses (or requests, in the ARP case).
>>
>> May be a good time to simplify.
>>>
>>
>> It seems like perhaps a little more hierarchy/coordination could result in
>> more simple operation at scale.
>>
>>
>>>
>>> Bob
>>>
>>> On Fri, Jul 7, 2023 at 2:25 PM Dave Taht via Make-wifi-fast <
>>> make-wifi-fast at lists.bufferbloat.net> wrote:
>>>
>>>> preferably using some sort of async circuitry for minimum interference
>>>> and power consumption. I figured, oh, 256MB would be more than enough
>>>> for a 10Gbit router.
>>>>
>>>> Instead, we can now layer 64GB on die.
>>>>
>>>> https://blocksandfiles.com/2023/07/05/3d-stacked-dram-and-processor-cube/
>>>>
>>>> --
>>>> Podcast:
>>>> https://www.linkedin.com/feed/update/urn:li:activity:7058793910227111937/
>>>> Dave Täht CSO, LibreQos
>>>> _______________________________________________
>>>> Make-wifi-fast mailing list
>>>> Make-wifi-fast at lists.bufferbloat.net
>>>> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>>>
>>>
>>> This electronic communication and the information and any files
>>> transmitted with it, or attached to it, are confidential and are intended
>>> solely for the use of the individual or entity to whom it is addressed and
>>> may contain information that is confidential, legally privileged, protected
>>> by privacy laws, or otherwise restricted from disclosure to anyone else. If
>>> you are not the intended recipient or the person responsible for delivering
>>> the e-mail to the intended recipient, you are hereby notified that any use,
>>> copying, distributing, dissemination, forwarding, printing, or copying of
>>> this e-mail is strictly prohibited. If you received this e-mail in error,
>>> please return the e-mail to the sender, delete it from your computer, and
>>> destroy any printed copy of it.
>>> _______________________________________________
>>> Make-wifi-fast mailing list
>>> Make-wifi-fast at lists.bufferbloat.net
>>> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>>
>>
>
>
-------------- next part --------------
_______________________________________________
Make-wifi-fast mailing list
Make-wifi-fast at lists.bufferbloat.net
https://lists.bufferbloat.net/listinfo/make-wifi-fast