MathJax CDN shutting down on April 30, 2017 | Hacker News Reader

MathJax CDN shutting down on April 30, 2017

mathjax.org

151 pointsJohnHammersley75 comments

Will the developers of Chromium and Edge perhaps reconsider incorporating native support for MathML in light of this, considering that apparently people do use maths on the web?

180 million monthly users is a significant portion of the world's population. I'm not sure to what extent MathJax is cached, so it's possible that these are not 180 million unique users. On the other hand, this does not include all those self-hosting MathJax, using KaTeX or just using server-generated images (for example Wikipedia†).

†It's possible to force Wikipedia to use native MathML (which, at least on Firefox, looks really nice) or MathJax, but it's non-trivial.

https://bugs.chromium.org/p/chromium/issues/detail?id=6606

https://bugs.chromium.org/p/chromium/issues/detail?id=152430

https://wpdev.uservoice.com/forums/257854-microsoft-edge-dev...

Yeah; I'm quite surprised, particularly given both Google's and Microsoft's push into the education market, that there is still not solution for this. MS certain has the edge (pardon the pun) with its maths tools in Word and Onenote, but lacking on the web. And that Google has given absolutely no consideration to this with Docs/Drive and related tools means that my school can't switch exclusively to G-Suite, and need to keep MS tools (and have to deal with installations and licensing) around.

People use math, but they use it with Latex not MathML, MathML is too verbose, so even if browsers implemented it people would need libraries like MathJax.

Nothing is stopping you from authoring your math however you like as long as there are tools that support the translation[1]:

[1]: http://fred-wang.github.io/TeXZilla/

I wish people used LaTeX.

I only get "what's that?" most of the time.

I don't think that LaTeX is a significant competitor of MathML, just like markdown isn't a competitor to HTML. If you're writing something for the web, you're highly unlikely to be directly inputting HTML, whether it's on Wikipedia, StackOverflow or Hacker News, so you'll need some sort of server-side processor or a javascript library previewer, anyway.

Native (browser) support of MathML is far more convenient for the reader and not more cumbersome for the writer.

It would be interesting to see what they tried to bill them. I can push out 70TB for about $700 right now (through our 11 PoP anycast network). AWS S3 would have cost $5,717.85 according to their calculator. My guess is the bill is somewhere in the middle.

If you don't need a fancy anycast network, you can buy a 1GbE pipe from Hurricane Electric for about $400/mo and push 192TB/mo through it. Actually they include that when you get their $400/mo server cabinets in the bay area. Probably not the best option ever, but I don't think S3 is 14.29x better than it. Dedicated server providers like OVH might be okay with it too, probably ask them first. But if you had 2-4 servers and round-robined them my guess is you would be fine. You could get one server in Europe and one in NA and GeoDNS them and get probably-good-enough close to what an anycast network can do. For ~$200/mo. Not too bad.

I've given my rant a million times so I won't do it again, but "The Cloud" heavily upcharges on bandwidth. Or they tell you it's "unlimited" (a word I've learned through experience to hate) then tell you "except for the kind of stuff you're doing" and send a bill or shut you down. If you've got to ship a ton of bandwidth, you should at least look into the alternatives.

If you really want to go future tech, look into something like IPFS (https://ipfs.io) - it was built for stuff like this and it's going to make a real dent in transit costs for heavy data sets, CDNs, package distribution and the like. I absolutely cannot wait for the distributed web.

> I've given my rant a million times so I won't do it again, but "The Cloud" heavily upcharges on bandwidth. Or they tell you it's "unlimited" (a word I've learned through experience to hate) then tell you "except for the kind of stuff you're doing" and send a bill or shut you down. If you've got to ship a ton of bandwidth, you should at least look into the alternatives.

It stuns me just how bad it upcharges and how much it could be hurting their business. I can imagine so many use-cases where reasonably priced bandwidth would drive consumption elsewhere. I use cloud storage for backups without worrying about the insane cost of a restore. I could send large video production files to cloud storage, use cloud compute to produce renders and then download the processed files again without paying a multiple of the compute cost on bandwidth.

To illustrate a bit on that video example, 4K 422 ProRes HQ is 424 GB/hr [0]. Assuming I have say 2h of footage I want processed, that's 848GB. Then if I need say 2h on 2x g2.8xlarge GPU instances to process the video, that's $2.78 * 2 * 2 = $11.12 for the entire process. Then I need to download that 848GB again, that's $76.32. The cost of retrieving the processed data is 7x the cost of processing it, that's insane.

[0]: https://ssl.apple.com/final-cut-pro/docs/Apple_ProRes_White_...

Interesting that the solution doesn't change the problem at all. As far as I can tell, cdnjs doesn't pay Cloudflare either. So all the traffic still flows through Cloudflare, unpaid. Cloudflare is listed as a sponsor for cdnjs. Guess it just keeps the list of who CF subsidizes shorter.

Cdnjs advertises Cloudflare to developers in its url, it's a significant difference.

Realistically, many legacy sites using this for rendering will just break and stay broken. Academic upload-and-forget posts, etc. Users may be unhappy but the overall volume of requests could drop.

Additionally, every sponsorship may have a fixed cost to someone in the organization, some manager who would rather read a 20 line vs. 100 line report. And I doubt mathematicians and physicists drive the CF bottom line. Generosity has its limits.

I thought the same thing regarding legacy sites. My workplace blocks file sharing and image hosting websites, so I've been acutely aware of the impact to UX recently when domains disappear. I can be almost certain now that when there are no images on a blog or stackoverflow post that they were hosted on imgur. I've been considering this for a while but it seems somewhat curious to me that not so long ago hotlinking was a big no-no and as sites disappeared broken image icons on small websites multiplied, and now we have an increasing number of CDNs where hotlinking is essentially the intended usage.

Have you asked what the justification is for allowing general internet sites but not image hosting sites? I can't imagine it's at all reasonable.

It's a fairly common rule in an enterprise. Unless you can guarantee that imgur will never host porn, it's going to get blocked per an HR policy.

But you can't guarantee that random blogs don't host porn either.

I believe it's mostly about legal liability vs actual effectiveness.

If someone watching porn causes someone else to file a hostile workplace suit, you can show that you invested in whatever is the standard off the shelf tech to prevent it.

It's all kind of silly today now that most people walk around work with 4g phones with high res screens. That leaves preventing malware on company PCs as the sole actual reasoning.

And why would the liability be different between random sites and imgur? I fully understand the idea of blocking both. I don't understand the idea of only blocking imgur.

It's meeting the legal definition of "reasonable care" to buy an industry standard blocking tool and maintain it with software updates.

The companies that use this software are checking boxes like "no porn", "no gambling". They aren't typically making URL level decisions. The stuff that sneaks by will happen because filtering isn't perfect.

The makers of the software categorize popular sites with tags (porn, file sharing, proxy avoidance, hate speech, etc). Less popular sites get content keyword filtering...so something like porn images without porn words can slip by.

This is a reasonable description of the internet filters we have. There is a lot more than just file sharing sites blocked, and they usually specify the category they are blocked under. The others just don't affect me as much. Though my above comment about IP paranoia still stands. We have other policies around access to email, USB permissions, bans of VM software etc purely to raise the bar above 'easy' leaks.

The genuine answer in many enterprises to something like that is "really? Time to block random blogs then".

Hilariously, EE Mobile in the UK does the reverse of this. Quite a lot of random blogs are blocked, but imgur and reddit are not blocked. Although if you're foolish enough to access reddit over http not https, certain subreddits are blocked.

Corporate systems don't care about a mythical "100% effective", they just want a porn blocking solution with acceptable false negatives. Which gets rid of imgur.

> Unless you can guarantee that imgur will never host porn

Enterprise policies don't need to be that enthusiastic before they'll block imgur. Imgur does in fact host porn.

Paranoia surrounding potential IP theft and leaks.

I never thought to archive all of Imgur for links from SO, but now I will! Thanks for mentioning this!

Common enough that countermeasures exist:

https://github.com/davisstu/imgur-to-filmot

(Chrome plugin that rewrites imgur links to point at some kind of imgur proxy that sits on a lesser known domain. No idea if this particular one works still.)

What a huge mess.

- Only 1 month to re-route 70TB of content. I wonder if even over 10% will be re-routed, with all the other sites broken.

- Cloudflare notice could have beem some months in. Hopefully now that this got some traction they will provide some more time.

- There are a lot of situations where devs cannot change the CDN any longer. I for one won't trust CDNs (including CDNJS) for anything durable any longer, which seems a big loss for Cloudflare.

Worse, mathjax is very popular with statically generated sites. Any pelican or Jekyll website you see with maths probably has this CDN hardwired into it. I am involved with 2 projects where we hard wired it into our docs.

Worse, the full download is massive, so it's tempting to not local host it. There are some cryptic messages about reducing the size, which link to out of date pages. If they were going to do this, they could at least have decent docs on how to locally host in a small size.

It's right in the article, under the "Optimization" header.

> I for one won't trust CDNs (including CDNJS) for anything durable any longer, which seems a big loss for Cloudflare.

Why? This has nothing to do with CDNs. This is a single account that can't afford to pay the bill anymore. CloudFlare itself isn't going anywhere.

You can use https://www.jsdelivr.com which is one of the best open-source hosts backed by several networks - or just host everything yourself and put a CDN in front for your own users.

Is there a way in which MathJax could redirect their cdn.mathjax.org to cdnjs.cloudflare.com and maintain backwards compatibility for existing sites?

A DNS redirect would break the SSL, right? Perhaps a HTTP 3xx redirect?

Yes.

HTTP 3xx would work fine, but you would need an active server there. But really setting up nginx with such rules as basically the only thing it is doing to handle these requests on OVH or some other fairly unlimited bandwidth server shouldn't be too hard.

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.

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

In 2011 were they piping 70TB?

You say "niche" but...

> 179 Million monthly users and 70TB traffic last month

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 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.

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?

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.

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.

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.)

I have an old web site hosted by my university that has several math equations on it using MathJax. Unfortunately, since graduating, I no longer have access to it, but my professor still hosts all of past students' content for us.

I'm just glad that I had the foresight to have hosted my own copy of that library.

I switched from MathJax to a middleman + execjs + KaTeX pipeline for my static sites earlier this year, and I'm very happy with it overall. KaTeX isn't 1-1 with MathJax when it comes to syntax support; I was surprised to see that underset and overset - which are ubiquitous - aren't supported. That said, I was able to work around every issue with alternative syntax. It's awesome that the heavy lifting is no longer client side, and it's a significantly easier build pipeline than MathJax if you're trying to generate a single minimized script or use webpack.

Is there a reason why ebooks don't use KaTeX? I subscribe to safaribooksonline.com, which presents HTML versions of ebooks. I think they're just unpackaged versions of epub or mobi versions. Anyway, I'm surprised that equations are generally embedded in images, even in recent (published this month) O'Reilly-published books.

Is there some technical reason why authors/publishers do it this way, or is it just about tooling?

Gitbook claims that it works: https://www.gitbook.com/blog/features/katex

Github code search for 'script src "cdn.mathjax.org"':

> 1,811,496 code results.

Woof.

It's kind of strange to think that you need fewer of your own server resources the more successful your software is at meeting its goals, but this seems to be one of those cases.

Is there any way to make Sphinx use its own copy of the script rather than the CDN?

I wonder if BuiltWith would be willing to release a list for free of sites using MathJax, so someone could build a tool to notify them all if contact info can be found

I just noticed this last week. I was trying to work out how to use MathJax directly from within JS to render math inline with rendering the rest of my content, rather than asynchronously after the fact.

I love MathJax, but I'm now rendering math on the server using a latex -> dvisvgm pipeline and caching the results

Depending on your use case, KaTeX [1] may be worth a look. It has a "renderToString" function that returns an HTML string.

[1] https://khan.github.io/KaTeX

One problem of KaTeX is that its feature set is too small. It doesn't even support some very basic commands such as \overrightarrow.

MathJax is such a great thing. Paired with simple markdown it really helps me structuring all my study work, sharing is so much easier with HTML files than TEX files. So sad to see the CDN go. Also: Hey Google, wanna do something really great? Sponsor them! wink

Hey Google, wanna do something really great? Sponsor them! wink

Better yet, add native MathML support to Chrome, and help eliminate the need for MathJax altogether. In the meantime, sponsoring MathJax would be a Good Thing though.

Jupyter Notebooks containing maths and exported as html are also going to get affected. I personally write my articles in a Jupyter Notebook and export them to html and put them on my site so need to update a few of them.

I thought cloudflare didn't have a data cap? I'm somewhat concerned.

See https://www.cloudflare.com/terms

Specifically, SECTION 10: LIMITATION ON NON-HTML CACHING.

Lots of products become popular when they are actually announced here as shutting down :)

I read that as MaxCDN shutting down.

> We recommend cdnjs which uses CloudFlare for delivery

Oh man, so close to

> We recommend cjdns (which is against everything CloudFlare stands for) for delivery

> which is against everything CloudFlare stands for

Could you explain this? Have I missed some big Cloudflare news?

For context, cjdns appears to be a routing suite for implementing a darknet overlay over the normal internet. I'm not entirely sure what Cloudflare is perceived to stand for, but it is a strongly centralized service and cjdns appears to be going for decentralization. (But I'd imagine a Cloudflare-like service, including centralized management = economies of scale in sysadminning, would still be valuable in a cjdns darknet.)

See https://github.com/cjdelisle/cjdns , https://hyperboria.net/ , and https://www.reddit.com/r/darknetplan/

cjdns is a different thing from cdnjs

Thanks, I thought it was a typo.

Reply on news.ycombinator.com