5 ms·

>The whole "edge compute" space seems poised to do well to me. IMHO, I think it's mostly a solution in search of a problem. The internet backbone is fast enou

by treis 8y ago

>The whole "edge compute" space seems poised to do well to me. IMHO, I think it's mostly a solution in search of a problem. The internet backbone is fast enough to not be noticeable for end users. A centrally located server in the U.S. will have a maximum ping of 40 ms to anywhere in the U.S. That's faster than is perceptible to the end user. The only mainstream usage I can see is cloud gaming where ping is that critical.

gmanley 8y ago
It may not make sense if all your customers are in the US but locations like Australia and others very much have an issue with latency. Especially if the server they are hitting is around the globe.
- treis 8y ago
  NYC to Sydney is still only ~200 ms. That's probably in the realm of being noticeable but only barely. Even then, the real need is more like one server per continent rather than every cell tower.
  - huac 8y ago
    200ms is human reaction time, you can definitely notice it; the rule of thumb is 100ms https://www.pubnub.com/blog/how-fast-is-realtime-human-perception-and-technology/ https://www.pubnub.com/blog/how-fast-is-realtime-human-perce...
    - treis 8y ago
      Your cite says that < 1s is "for users to feel they are interacting freely with the information". Even a 400ms round trip is well within that limit.
      - sajal83 8y ago
        Most, if not all websites require more than one round trip.
        wbl 8y ago
        HTTP2 can reduce this significantly.
  - bogomipz 8y ago
    That statement is categorically false. One simple example is online gaming/MMOGs where a latency of 200 ms would be be a miserable experience. The upper bound often targeted for "acceptable interactivity" is somewhere between 100-130ms.
    - treis 8y ago
      (1) I excluded gaming as something that is legitimately ping dependent (2) You can achieve 100-130 ms latency with a "like one data center per continent" sort of network. You don't need edge computing on hundreds or thousands of edge sites to achieve that.
      - cagenut 8y ago
        I'm sorry but what are you talking about? Fastly has about 60 PoPs (physical locations). No one is talking about hundreds of thousands of sites. Even akamai, who made it one of their primary marketing metrics, only ever got into the thousands.
        treis 8y ago
        The GP is talking about putting them in cell phone towers.
  - cagenut 8y ago
    Your numbers are wrong but your line of reasoning is actually the basis of how Fastly's network was designed to be different from Akamai's. There's a graph about it at the 2:40 mark in this video: https://vimeo.com/132842124 https://vimeo.com/132842124
    - lapnitnelav 8y ago
      That presentation was refreshingly honest and interesting to boot with.
stickfigure 8y ago
See also: Page takes 2500ms to render javascript.
nyc_pizzadev 8y ago
That's only good for small media like websites and images. Videos and downloads need to be cached in the provider networks, you simply cannot serve terrabits of data from a single origin. Well, you can, but networks don't like to operate like that. So ya, at scale, you need edges. The internet is dominated by companies at scale.
- treis 8y ago
  That's more of a CDN than edge compute. Edge compute would be something like having your Rails app run in a bunch of edge locations and then doing some eventually consistent magic to sync them up.
bogomipz 8y ago
>IMHO, I think it's mostly a solution in search of a problem. I agree, I've been looking at discussion one here recently on how someone was working on the problem of "data storage at the edge" to compliment their "compute at the edge" offering. I think you can argue that that when you move compute and data storage to the edge you've effectively moved the data center to the edge. At some point if you put enough data at the edge you operating your own backbone makes economic sense and CDN don't maintain backbones they mostly buy paid peering and transit and try to leverage public peering wherever they can. I would even say edge compute is a solution in search of revenue stream and hedge in a heavily commoditized market.
merreborn 8y ago
https://hpbn.co/transport-layer-security-tls/#leverage-early-termination https://hpbn.co/transport-layer-security-tls/#leverage-early... Establishing a connection can require several round trips, and the latency adds up. "early termination" with a point of presence near your end user makes a pretty massive difference, especially if your user isn't on the same continent. Every additional 100ms costs amazon 1% of their revenue https://news.ycombinator.com/item?id=273900 https://news.ycombinator.com/item?id=273900
paxys 8y ago
Gaming is a bad example because it isn't the ping to the edge server that matters but ping to another user.

by treis 8y ago

gmanley 8y ago

It may not make sense if all your customers are in the US but locations like Australia and others very much have an issue with latency. Especially if the server they are hitting is around the globe.

treis 8y ago
NYC to Sydney is still only ~200 ms. That's probably in the realm of being noticeable but only barely. Even then, the real need is more like one server per continent rather than every cell tower.
- huac 8y ago
  200ms is human reaction time, you can definitely notice it; the rule of thumb is 100ms https://www.pubnub.com/blog/how-fast-is-realtime-human-perception-and-technology/ https://www.pubnub.com/blog/how-fast-is-realtime-human-perce...
  - treis 8y ago
    Your cite says that < 1s is "for users to feel they are interacting freely with the information". Even a 400ms round trip is well within that limit.
    - sajal83 8y ago
      Most, if not all websites require more than one round trip.
      - wbl 8y ago
        HTTP2 can reduce this significantly.
- bogomipz 8y ago
  That statement is categorically false. One simple example is online gaming/MMOGs where a latency of 200 ms would be be a miserable experience. The upper bound often targeted for "acceptable interactivity" is somewhere between 100-130ms.
  - treis 8y ago
    (1) I excluded gaming as something that is legitimately ping dependent (2) You can achieve 100-130 ms latency with a "like one data center per continent" sort of network. You don't need edge computing on hundreds or thousands of edge sites to achieve that.
    - cagenut 8y ago
      I'm sorry but what are you talking about? Fastly has about 60 PoPs (physical locations). No one is talking about hundreds of thousands of sites. Even akamai, who made it one of their primary marketing metrics, only ever got into the thousands.
      - treis 8y ago
        The GP is talking about putting them in cell phone towers.
- cagenut 8y ago
  Your numbers are wrong but your line of reasoning is actually the basis of how Fastly's network was designed to be different from Akamai's. There's a graph about it at the 2:40 mark in this video: https://vimeo.com/132842124 https://vimeo.com/132842124
  - lapnitnelav 8y ago
    That presentation was refreshingly honest and interesting to boot with.

stickfigure 8y ago

See also: Page takes 2500ms to render javascript.

nyc_pizzadev 8y ago

That's only good for small media like websites and images. Videos and downloads need to be cached in the provider networks, you simply cannot serve terrabits of data from a single origin. Well, you can, but networks don't like to operate like that. So ya, at scale, you need edges. The internet is dominated by companies at scale.

treis 8y ago
That's more of a CDN than edge compute. Edge compute would be something like having your Rails app run in a bunch of edge locations and then doing some eventually consistent magic to sync them up.

bogomipz 8y ago

>IMHO, I think it's mostly a solution in search of a problem. I agree, I've been looking at discussion one here recently on how someone was working on the problem of "data storage at the edge" to compliment their "compute at the edge" offering. I think you can argue that that when you move compute and data storage to the edge you've effectively moved the data center to the edge. At some point if you put enough data at the edge you operating your own backbone makes economic sense and CDN don't maintain backbones they mostly buy paid peering and transit and try to leverage public peering wherever they can. I would even say edge compute is a solution in search of revenue stream and hedge in a heavily commoditized market.

merreborn 8y ago

https://hpbn.co/transport-layer-security-tls/#leverage-early-termination https://hpbn.co/transport-layer-security-tls/#leverage-early... Establishing a connection can require several round trips, and the latency adds up. "early termination" with a point of presence near your end user makes a pretty massive difference, especially if your user isn't on the same continent. Every additional 100ms costs amazon 1% of their revenue https://news.ycombinator.com/item?id=273900 https://news.ycombinator.com/item?id=273900

paxys 8y ago

Gaming is a bad example because it isn't the ping to the edge server that matters but ping to another user.