4 ms·
As someone who works for a CDN (not CloudFlare), this was certainly interesting to read. It is always good to remember that we owe something to our customers' e
by cortesoft 7y ago
As someone who works for a CDN (not CloudFlare), this was certainly interesting to read. It is always good to remember that we owe something to our customers' end users in addition to our direct customers, the website owners.
We have to remember that those are real people and not just percentages of traffic that are affected when we make decisions like putting a captcha up against every visitor from certain countries. I like to think we were already doing that, but it is good to be reminded.
However, I am not quite sure I am getting what the author is suggesting when they say that sites should forgo a CDN. Maybe I am biased, but if you thought latency was bad when a datacenter near you went down for maintenance, try having to go all the way to the sites origin in New Jersey for every request. I am not aware of any way besides a CDN (or a CDN like setup) that would get you good performance for people in all countries.
So I get the frustration with the captchas, and I get the frustration with the lack of multiple datacenters near you, but I wonder if you will make things worse for yourself by advocating to not have a CDN.
- derefr 7y agoIn the author’s case, the CDN was worse than fetching from the origin; for some reason, Cloudflare just gave up on rerouting their traffic (not sure how this is even possible given their anycast DNS) and they never got the requested resources at all! Having even a 30s RTT would have actually been an improvement vs. this outcome.
- cnst 7y agoIf you have a proper website without a gazzilion of useless JavaScript frameworks and custom fonts, there's hardly going to be an unacceptable experience latency-wise even if you have to go all over the world to fetch those resources.
- cortesoft 7y agoI guess I didn't realize, until another commenter replied, that their traffic was actually blackholed... I thought it just had really bad latency, which they were complaining about.
- toast0 7y agoA working CDN certainly helps performance a lot, but it's not impossible to have acceptable performance worldwide from US only hosting. The three big things you need to do are: A) serve your site quickly --- there's going to be a lot of not as avoidable latency between you and the user, but anything after your server gets the request is on you. B) Keep your page weight small. Of course, this is a good idea anyway, but transfer rates on high latency connections are more often limited by tcp slow start than bandwidth. C) Use TLS 1.3 or at least TLS 1.2 with ALPN (which triggers TLS false start in at least Chrome, and I think other browsers), as these reduce the number of round trips during connection setup compared to standard TLS 1.2 or earlier. It's worth measuring http/1.1 vs http/2 vs http/3 to see what works best for a particular site on high bandwidth networks. Honorary mention: make sure path MTU blackhole discovery is working. There are still plenty of networks that have path MTU blackholes, and sending packets the size clients said they could get doesn't always work.