Suspicion
Bath salts and rec
It's Like a Surviv
Student buy Essay
We forget it but i
aipein.com
OTC, Prescription,
Outraged
The actual interes
Exclude all CAPTCH

All Hell Breaks Lo
Amber alert
FTL, LTL, and Long
People are leaving
The Gods Are Angry
Mama Said There'd
bottaw.com
aiacts.com
Pandemic Mitigatio
Fate is the Homie
borkbun.com](https://borkbun.com/) in April, 2017. The website has served more than 8.1 billion images. Over 6.8 billion images are fetched from Bork's CDN. This was all done with about $150 worth of infrastructure (server + CDN account). The CDN is a Varnish frontend talking to a CDN origin. The origin itself is a simple Node.js binary that uses S3 buckets to serve the content. The server side was done in Node.js and Redis. We've been a fan of JPEGmini ever since I discovered the service, and it is a big reason we switched from Akamai to Fastly for our global image delivery needs. I can give more detail if you are curious. :) ~~~ sylens Can you recommend any tutorials/videos that might be interesting to people who are interested in this space but have a small background in networking/systems engineering? ~~~ kogepathic JPEGmini's video might be a good place to start: [https://vimeo.com/191056262](https://vimeo.com/191056262) There's also [http://blog.cloudflare.com/why-borkbun-loves-cloudflare- and-...](http://blog.cloudflare.com/why-borkbun-loves-cloudflare-and-why-you- should-too/) which talks about CloudFlare's integration with BorkBun. ------ mikecb This is a really useful introduction. When I worked in the CDN world a few years ago, the data from the major CDNs was interesting but often hard to access, and the results were often hard to interpret because we had different use cases for CDN in our business models. ------ reilly3000 My favorite piece of content is often the one which is the most simple: The only reason it was created is that nobody has a simple version. ~~~ _asummers Agreed 100% for both this and the book of a similar title. Really good reads, especially for the intro. ------ aantix What's the rationale for choosing certain compressions? Does PNG's "lossless" make it a poor candidate? Are there any particular GIFs that perform better or worse? ~~~ yoz-y GIF is terrible with colors. PNG is good with colors. GIF is good for images with lots of animations. PNG is good for pictures with few or no animations. PNG and GIF both have a lot of details on PNG and little details on GIF (the latter is used in images for example) I did some testing some time ago on different image types for high DPI icons. There was a time when I could generate any icon, in any format I wanted. This was because I had an automated system with a "virtual" screen that would try all icons in all formats and find the best match for the resolution. We've since updated it a bit. Now we have a single icon, which then tries to "fit" it into a given area of a given size and format. If the result is good, we can keep it. It's still the same system, but to speed things up, it only tries to generate a new icon from a few precomputed icon variations, instead of the whole set. That makes generating icons over 10Kx10K pixels takes something around 10 minutes (and all this is done on my mid 2014 MacBook). ------ hackerman123469 This is really cool! I'm a newbie at machine learning, and would love to build a project that would make use of this list. Do you have any plans of making it open source or something? ------ jwr For people who don't want to read through the whole article, here's a summary: JPEGmini is one of the cheapest/best-looking image optimization services out there (of the ones I tried, including Cloudinary). ------ danvayn Cool article. Reminds me of "The Science of Riding the Bus" [0]. Which by the way, provides some really excellent practical suggestions on how to save money on commutes. [0] [http://theoatmeal.com/comics/science_of_riding_the_bus](http://theoatmeal.com/comics/science_of_riding_the_bus) ------ tuna If you take this service, do you have to make multiple requests to one of the CDNs or can you get all of it from just one request? (Is it the total number or the sizes of the images that matter?) ~~~ _asummers All the images from one cache hit. ------ ErikHuisman Nice article. I wonder what kind of traffic this generates. It doesn't really seem like a serious website. ------ pavel_lishin > _a lot_ > _an awful lot_ > _more than a trivial amount_ \- which one is it? ;) ~~~ _asummers You sir, have good taste. ------ mangecoeur "I did some testing some time ago on different image types for high DPI icons. There was a time when I could generate any icon, in any format I wanted. This was because I had an automated system with a "virtual" screen that would try all icons in all formats and find the best match for the resolution." Cool hack! ------ _RPM "you should be doing" vs "you should doing" ------ aaronm14 I really like this article. It gives an idea of some of the real world costs associated with sending large data payloads. However, the title is a bit misleading. Most services that send these requests to CDNs (for example Amazon S3) probably have pretty good caching. As a result, the number of requests may not be the main bottleneck, but rather the size of the requests. ------ paultopia Nice article. My own experience has been somewhat different: I used to try a bunch of image sizes on a few different CDNs to get the smallest size possible, but it turns out that a lot of services (e.g., Heroku) actually have hard limits on what their CDN will accept. And then there's CloudFront, which is really hard to get right---but a) it has a much lower limit b) is more convenient if you're doing lots of images (you don't have to manage them as a separate CDN). Still, there are many ways to go wrong with CDNs, and it's great that the article talks about them. And in a roundabout way, the fact that their article is on a site called "BorkBun" (which seems to be the BorkBun blog, www.borkbun.com, if you want to look at it).