Hacker News Reader

Comment by joshmn | Hacker News Reader

So, yeah, ugh, why would a niche library (in the sense that it's not jQuery or YUI) need to operate its own CDN, for anything other than style points? There are major players who are already doing this who own and operate their own network (see MaxCDN). What's wrong with just plugging into their existing infrastructure?

If that fails, I'd give a shout to FDCServers or LeaseWeb (or of that capacity) and see what they can do. I know of a handful of people who've moved 100TB monthly for less than $100/month, just because they asked.

View on HN

Why is this downvoted? Nitpicking of the phrase "niche library" aside, I'm having trouble understanding this without more context as well.

The post starts:

> Our CDN has been an important part of MathJax’s history. When MathJax made its first public release in 2010, hosting a library like MathJax was a complex challenge. The CDN launched a year later and helped resolve this difficulty

What is "like MathJax" alluding to? Standing up a web host to serve a single JS file sounds like the opposite of a "complex challenge", so there must be something else going on here.

After reading further down, I see that the post does mention that "A common concern with local installations is the size of a complete copy of MathJax, in particular the ~29,000 PNGs", so I guess that's what they're talking about? If so, the comments at the top make more sense and their having had their own CDN at all is more understandable in that light, but it seems needlessly confusing not to lead with that explanation.

Your guess is as good as mine. I edited my original post to reflect my subjectiveness of niche.

I guess my point was that... Okay, so they took advantage of CloudFlare. Great, plenty of people do that, but not on the scale of 10s of TBs. What plan were they on with CF? The $20/month one? If I was a provider, I wouldn't want someone moving that volume of data through my pipes for $20/month (but as a developer, I'd love it)

They mentioned how Google provided free storage. Storage is so cheap! Providers make their money on bandwidth. Having said, I can't imagine what kind of price they paid for bandwidth.

Like I alluded to in my (now edited) original post: Guys, it never hurts to do your research in the hosting space and learn that there is more to hosting than just AWS/Google/Azure/Heroku/OVH/DigitalOcean/Vultr. Go to webhostingtalk.com, learn a bit, browse the offers forums, write some emails, compare providers, ask questions. Hosting isn't complicated, especially compared to some of the problems you're/we're solving.

Let's say you don't care about DDOS protection and can live with 0.1x the number of PoPs. How much does it cost these days to host these terabytes/month from ≥10 well spread locations?

If you want to maintain the basic benefit of a CDN, the cheapest I know of is OVH's CDN. MathJax says they use 70TB/month.

The OVH 100TB CDN plan is £509.99/month. Other CDNs would be 3x that amount or more.

It wouldn't be available much cheaper even if you skipped the CDN. A $20 VPS comes with 3TB outbound, so you would need 24 of them to spread across.

Of course, there may be some opportunity to fix their cache headers, or optimize their js/CSS/images and reduce that 70TB requirement.

> A $20 VPS comes with 3TB outbound

You can get $40 servers with 30TB outbound.

I assume that's a 100mbps unmetered port, and probably not able to push 100% at peak times (over provisioned). How many of these would you need to serve the implied peaks that come with 70TB/month? And have them spread on at least 2 continents?

Gigabit port limited to 30TB. It should be able to handle the load patterns fine. That's from Hetzner. That's only one continent by itself, but that's a very significant chunk of the load solved with 1-2 servers.

If you're willing to stretch the budget a little, many OVH servers actually come with 500 mbps guaranteed.

I have one that I pay $80 a month for and push 30TB monthly with no issues.

Only issue with OVH is that you aren't going to get the geographical coverage you would with a proper CDN like Cloudflare. But if you're on a budget, it's pretty good.

If you care about cost you can get away with only a couple of locations, totaling maybe slightly over $100 monthly.

I don't see 70TB of outbound bandwidth fitting into $100/month.

Even with unmetered bandwidth servers, you would need quite a few of them. 100Mbps, full throttle, for a month, is less than half their 70TB/month requirement. And that's if the bandwidth needs were even across the whole day, without spikes.

Even a bare bones budget setup is going to be $300-$400/month, and that's cutting it very close and providing crap service with some users downloading from across the globe.

OVH or Online.net alone can get you that much bandwidth with a dedicated server and for less money. Use one to serve US users and the other one to serve EU users. CDNs are not a zero-sum game, you will always have users downloading "from across the globe" or over slow connections.

I'm assuming a budget provider like OVH or online.net to get to the $300-$400 a month already. I suppose, yes, you could instruct users to pick from a North American or EU set of urls.

"I suppose, yes, you could instruct users to pick from a North American or EU set of urls."

No, you instruct your DNS to point to one location for EU users and another location for US users, urls don't change.

Have you seen Cloudflare map? 100 locations across all inhabited continents, even Africa gets a few...

And growing... we want to be 10ms from the entire online population.

FDC comes out to ~$350/month. They always have promos going on, wouldn't be surprised to that number get halved without much effort.

I don't think it's a single JS file. The query string seems to be significant. I currently request MathJax.js?config=TeX-AMS-MML_HTMLorMML. (I set it up years ago, so I have no particular recollection of choosing that query string, but I'm confident it was significant.)

You say "niche" but...

> 179 Million monthly users and 70TB traffic last month

Was this the case in 2011 when MathJax began offering this? I don't know the answer.

In 2011 were they piping 70TB?

Reply on news.ycombinator.com