Hacker News Reader

Comment by pjf | Hacker News Reader

The site seems slow/down now. A cached copy is available, obviously, on Google :)

https://webcache.googleusercontent.com/search?q=cache:m-ItC3...

I asked myself, why on earth do websites have to be slow when they're under heavy load?

To be fair, the technical reason is that the modern web (circa last 3-5 years or so?) has mostly switched over to demand scaling, and this is why hug-of-death situations are comparatively rare now.

If you have to be super-economical, and even setting an upper spending limit is uncomfortable, set your site up to ping you when/if it's falling apart so you can pull the trigger on ordering more cores yourself. Maybe put a super easy "order more cores" button in the ping, make 2:30AM disasters less terrible.

Of course, technical reasons aside, my initial unimpressedness (or, I must admit, indignation - ha) was directed at whoever this group's ISP is. Clearly the ISP is not providing enough [demand] bandwidth.

[Edit: After reading some comments below I think I should insert a preface/prefix here and say that it's entirely possible the website itself is misconfigured. I think I got sufficiently indignant that I didn't factor this in, and this was definitely an oversight. I don't know the situation, and it's unfair to squarely blame the ISP. Original message continues unedited...]

So, FWIW, this group is using mediatemple. (Found by a simple google of the IP and careful untangling of the results.) ...Well that's a name I've not heard in years. And it made me wonder - do MT have an answer to demand scaling? I'm not able to discern either way from a quick browse of their website (including the "managed wordpress hosting" section). If they don't (and they seem to be stuck in the VPS days), ouch. If they do have demand scaling, then I think it's unprofessional to offer configurations that don't handle situations like this.

In any case, a "504 Gateway timeout" with a tiny "nginx" underneath is going to make people go "hmm, who's this person's ISP" and knock a few points off whatever the answer to that question is. I wouldn't want that happening to me or anyone I build websites for.

[Edit, in the same vein as the edit above: the conclusion above is plausible, but incomplete. Again, it's necessary to consider all angles.]

I'm always wondering the same. As a pretty noob sysadmin, my dead cheap[0] second-hand dedicated server was able to handle being on the HN homepage for 11ish hours without breaking a sweat. No CDN was involved what so ever, all content was going directly to my single point of failure.

Yet, I have to resort to looking at cashed versions of the submissions from companies that have way more resources than I could ever imagine to have.

[0] Price of the cheapest Media Temple VPS offering.

Hrm. Very well put :) thankyou for this perspective. Highly appreciated.

I hadn't really factored in website optimization. Who knows if each page load is chewing CPU for a couple of seconds because of an unoptimized query. I was only reading the other day about a random situation along those kinds of lines...

Looking at this page's source it seems like Wordpress (or rather Shitpress) is involved, so this downtime is not surprising at all considering how many resources that pile of garbage uses.

But yeah, a properly optimized site can handle being on the HN front-page on a single server with no issues.

There's plenty of caching plugins that make Wordpress run just fine at higher traffic. This site likely just doesn't have one installed.

Would you buy a new car if it needed duct tape just so it can stay in one piece while driving on the highway? So why would you think it's acceptable for a CMS to require third-party "cache" plugins just to be able to handle traffic?

Computers and cars are both solutions to problems, but the complexity of computers is yet to be routinely contained without targeted effort. This is especially true of the Web. Wordpress is the single most installed CMS (and kitchen sink, at this point), it's a massive attack target, and it's installed by people who don't really know what they're doing.

Between being a kitchen sink (and having to deal with the internal/external API/architectural [compatibility] baggage that comes with any long-lived implementation) and having to be the CMS equivalent of Windows/macOS, it's a... difficult ask for a WP installation with (say) 30 poorly-constructed addons to to run smoothly. Possible, yes, absolutely, but it'll need tuning.

Caching is simply the simplest bandaid that can be tacked on. If the caching is tuned properly, it'll fix everything else.

Yes, this will mean the underlying configuration/state will be the digital equivalent of a giant wound you just want to clean out and fix up, but, err... it works.

The vast majority of WordPress-powered sites have no personalised content: the only authenticated users are 1-2 administrators, and 1+ people who add/edit content.

For this use-case, it's hard to beat WordPress for features and ease-of-use, especially when you consider the ecosystem of plugins that can add commonly-needed functionality. Sure, you could roll-your-own, and then add some code to generate a sitemap.xml etc., but there's no need.

Sure, it doesn't scale without adding caching, but it's trivial to add caching with either a plugin, or by putting varnish or similar in front on the same machine, or using a CDN.

Sure, if you have dynamic or personalised content, you'll need to cache further down the stack (e.g. at the database level), and that's less efficient (because you're probably running multiple php files to generate a single HTTP response), but that's not the usual situation.

That analogy really has no real relevance to the discussion as far as I can see. Caches help dynamic sites scale up to meet large amounts of traffic. If you don't want to use a WP plugin then use Varnish instead.

For 99% of sites, a simple CDN solves all scaling challenges. And it's cheap. Even a 1-second cache timeout can drastically reduce the biggest load spikes.

I always load test my webpages with https://ddostest.me/load-test/ before advertising them on HN. Strangely, I often find a little something missing.

Nice article worth reading even though I got a few 504. It looks more professional when the webpage loads in a second on first try.

Reply on news.ycombinator.com