[HN Gopher] Skypack - A new kind of JavaScript delivery network ___________________________________________________________________ Skypack - A new kind of JavaScript delivery network Author : bryanbraun Score : 19 points Date : 2020-07-13 21:31 UTC (1 hours ago) (HTM) web link (www.skypack.dev) (TXT) w3m dump (www.skypack.dev) | lilyball wrote: | Does this support subresource integrity? | billyhoffman wrote: | They claim a high availability, but no data on P90 or P95 | response times. There is no SLA on their free plan. | | While I appreciate the free service, I work in the web | performance space. I've seen third party widgets from big name | companies like Optimizely, Tealium, Pubmatic, and Adobe have slow | response times for critical content, delaying the entire | waterfall for major sites. Its hard for the big companies to whom | you actually pay money to get this consistently right in a way | that doesn't slow down the page. | | We've had a ton of people taking cracks at this. From Google and | Microsoft with their "Ajax" CDNs a decade ago, to unpkg and | Cloudflare. And yet I still have not seen any of these special | JavaScript-specific CDNs ever beat a vanilla CDN dumbly serving | JS packages that you yourself have bundled and optimized under | normal site traffic. | ss3000 wrote: | Yeah, I think a service like this might be very useful for | quick prototyping and proving out an MVP thanks to the native | es6 modules, but I personally wouldn't rely on it for anything | critical. | | Coincidentally jspm.dev, a similar ES modules CDN in this space | recently released a new version as well: https://jspm.org/jspm- | dev-release | | Their release post had a lot more technical details (including | their strategy for code splitting which is critical for the | production use case) so I'm inclined to trust them a little bit | more. Curious where I could read more about the architecture | behind Skypack? | 1f60c wrote: | I thought "hmm, this seems _a lot_ like Pika[0]! ", only to find | out it _is_ Pika: | | > Skypack (Previously: Pika CDN) has already served over | 2,000,000+ JavaScript packages across 4,000 unique domains, | powering websites around the world. [0]: | https://www.pika.dev/ | pier25 wrote: | So once there are bundler plugins, what is going to happen here? | | You will deploy ES5, ES6, and ESM and then Skypack will decide | what to deliver to the browser? | flaque wrote: | Why the rename? | 1f60c wrote: | I'm curious about this as well, and the search box still refers | to Pika.dev | cosmotic wrote: | Oh, it's modern? What does that mean =/ | fxtentacle wrote: | What they fail to mention is that you lose the ability to remove | unused code, for example when you would otherwise import only | some Bootstrap components' JS. | | So in the end, their hosted copy might be much larger than if you | had hosted an optimized copy yourself. | | Using their service, it is also impossible for you to merge small | dependencies and your application logic into one file to reduce | the number of HTTP requests. The resulting higher number of gets | might well make it slower than not using their CDN in the first | place. | blaisio wrote: | You lose the ability to remove unused code, but for people | using this, it doesn't matter, because it is cached anyway. | | The number of HTTP requests really does not matter much these | days, because of HTTP2. And again for people who use this | service and get a speedup from it, it doesn't matter because it | is cached. | Quekid5 wrote: | > You lose the ability to remove unused code, but for people | using this, it doesn't matter, because it is cached anyway. | | I was under the impression that parsing (etc.) time matters | as well, not just raw download time/latency? | blaisio wrote: | Yes that's fair, with less code you have less to parse. | Parsing time is ideally a negligible amount of time, | especially because it can normally happen asynchronously, | but for very large libraries I could see it being an issue. ___________________________________________________________________ (page generated 2020-07-13 23:00 UTC)