5 ms·

What is edge compute?

32 points by kiyanwang 4y ago

tptacek 4y ago
This is a lot. "Edge compute" is compute (dynamic stuff; actual code running remotely, as opposed to static content) positioned close to users rather than in us-east-1 ("close" meaning "asymptotically approaching the user's cable modem", but in the main usually meaning "in a city close to the user").
- infogulch 4y ago
  > "Edge compute" is compute positioned close to the user, asymptotically approaching the user's cable modem. Nice description. 2k words summed up in 15.
honkler 4y ago
a buzzword
makz 4y ago
Normal compute that has existed forever.
- moffkalast 4y ago
  You try getting investor funding with "normal compute" haha
  - kevml 4y ago
    That feels close to the mission of https://oxide.computer/ https://oxide.computer/
    - moffkalast 4y ago
      Box 3 Gavin Belson Signature Edition? :D
jandrewrogers 4y ago
Applications having high statefulness is not a bad edge use case, it is often the reason why substantial edge deployments are done! I've typically seen the term "edge compute" to denote a deployment that solves a networking constraint which makes use of a traditional data center prohibitive: available bandwidth too low, latency is too high, or network is too unreliable. It is often cheaper to make the edge heavily stateful than to add sufficient network capacity so that all the data can be backhauled to a data center. Three reasons why there is often substantial state at the edge: you are doing sensor processing for operational reasons that intrinsically requires a lot of state to extract features of interest locally before sending those features upstream; you are aggregating feeds for operational reasons at the edge from various sensor platforms also at the edge to apply analysis across them; the edge platforms need to retain data for a specified period of time for regulatory, auditing, forensic, etc reasons. Not every edge case is like this and my sample is biased, but beefy boxes at the edge to avoid data backhaul are not that uncommon.
jaredcwhite 4y ago
I have to admit a certain personal bias coming into this, but honestly it's not hard to see how so much of this conversation is being driven by particular hosting companies and their pet frameworks trying to capture marketshare. IMHO: Most web projects don't need to be run "right on the edge". Most web projects don't need to contort themselves to fit into the category of "serverless functions" and all the limitations therein (and the edge compute platforms are even more constrained!). Many articles, like this one, completely gloss over the wonderful category that is "managed platforms which still give you real VMs". Take Render for example. (And platforms like Heroku before that.) I certainly don't need to deal with the headaches of upgrading OS versions or applying security updates or doing a ton of manual setup any time I need to scale to more servers. Render does that all for me automatically, or at the click of a button. Yet at the end of the day, I'm just writing regular server code to deploy onto "regular" Unix-y environments which are fairly similar to what I develop on at home. Pretty awesome stuff. Do I care that it's not all running "on the edge"? Nope. So while it's "cool" that we have things like edge compute as an option, I think the whiz-bang-ness of it all is being wildly oversold. Industry-wide, I want to see managed VM platforms like Render (and even lower-level platforms like, say, Fly) take the lead in people's minds rather than edge compute which is far more limited and—most damning—JavaScript-centric.
- qudat 4y ago
  Agreed. Deploying apps at the edge is a shiny new toy that companies with good marketing teams are pushing as the next big thing. Most products or services don’t need it. We will eventually find a good niche for edge computing instead of it being marketed as a thing everyone should use.
kkfx 4y ago
Far simpler: edge computing is the admission that modern mainframes, the cloud, are untenable, we need local ones. But since few giants do not want giving users ANY IT power and remain in control they offer a middle ground: you buy devices we manage and control, such device act locally as needed, pushing just something to us for our surveillance, in exchange you get "magic services", breadcrumbs of what you potentially can get but since you do not know just obey and be happy, we are the expert you are just a cow with only two legs. That's IME the simplest definition of edge computing...

32 points by kiyanwang 4y ago

This is a lot. "Edge compute" is compute (dynamic stuff; actual code running remotely, as opposed to static content) positioned close to users rather than in us-east-1 ("close" meaning "asymptotically approaching the user's cable modem", but in the main usually meaning "in a city close to the user").

infogulch 4y ago
> "Edge compute" is compute positioned close to the user, asymptotically approaching the user's cable modem. Nice description. 2k words summed up in 15.

honkler 4y ago

a buzzword

makz 4y ago

Normal compute that has existed forever.

moffkalast 4y ago
You try getting investor funding with "normal compute" haha
- kevml 4y ago
  That feels close to the mission of https://oxide.computer/ https://oxide.computer/
  - moffkalast 4y ago
    Box 3 Gavin Belson Signature Edition? :D

jandrewrogers 4y ago

Applications having high statefulness is not a bad edge use case, it is often the reason why substantial edge deployments are done! I've typically seen the term "edge compute" to denote a deployment that solves a networking constraint which makes use of a traditional data center prohibitive: available bandwidth too low, latency is too high, or network is too unreliable. It is often cheaper to make the edge heavily stateful than to add sufficient network capacity so that all the data can be backhauled to a data center. Three reasons why there is often substantial state at the edge: you are doing sensor processing for operational reasons that intrinsically requires a lot of state to extract features of interest locally before sending those features upstream; you are aggregating feeds for operational reasons at the edge from various sensor platforms also at the edge to apply analysis across them; the edge platforms need to retain data for a specified period of time for regulatory, auditing, forensic, etc reasons. Not every edge case is like this and my sample is biased, but beefy boxes at the edge to avoid data backhaul are not that uncommon.

jaredcwhite 4y ago

I have to admit a certain personal bias coming into this, but honestly it's not hard to see how so much of this conversation is being driven by particular hosting companies and their pet frameworks trying to capture marketshare. IMHO: Most web projects don't need to be run "right on the edge". Most web projects don't need to contort themselves to fit into the category of "serverless functions" and all the limitations therein (and the edge compute platforms are even more constrained!). Many articles, like this one, completely gloss over the wonderful category that is "managed platforms which still give you real VMs". Take Render for example. (And platforms like Heroku before that.) I certainly don't need to deal with the headaches of upgrading OS versions or applying security updates or doing a ton of manual setup any time I need to scale to more servers. Render does that all for me automatically, or at the click of a button. Yet at the end of the day, I'm just writing regular server code to deploy onto "regular" Unix-y environments which are fairly similar to what I develop on at home. Pretty awesome stuff. Do I care that it's not all running "on the edge"? Nope. So while it's "cool" that we have things like edge compute as an option, I think the whiz-bang-ness of it all is being wildly oversold. Industry-wide, I want to see managed VM platforms like Render (and even lower-level platforms like, say, Fly) take the lead in people's minds rather than edge compute which is far more limited and—most damning—JavaScript-centric.

qudat 4y ago
Agreed. Deploying apps at the edge is a shiny new toy that companies with good marketing teams are pushing as the next big thing. Most products or services don’t need it. We will eventually find a good niche for edge computing instead of it being marketed as a thing everyone should use.

kkfx 4y ago

Far simpler: edge computing is the admission that modern mainframes, the cloud, are untenable, we need local ones. But since few giants do not want giving users ANY IT power and remain in control they offer a middle ground: you buy devices we manage and control, such device act locally as needed, pushing just something to us for our surveillance, in exchange you get "magic services", breadcrumbs of what you potentially can get but since you do not know just obey and be happy, we are the expert you are just a cow with only two legs. That's IME the simplest definition of edge computing...