Tuesday, July 26, 2016

Jupyter: "take the domain name down immediately"

The Jupyter notebook is an open source BSD-licensed browser-based code execution environment, inspired by my early work on the Sage Notebook (which we launched in 2007), which was in turn inspired heavily by Mathematica notebooks and Google docs. Jupyter used to be called IPython.

SageMathCloud is an open source web-based environment for using Sage worksheets, terminals, LaTeX documents, course management, and Jupyter notebooks. I've put much hard work into making it so that multiple people can simultaneously edit Jupyter notebooks in SageMathCloud, and the history of all changes are recorded and browsable via a slider.

Many people have written to me asking for there to be a modified version of SageMathCloud, which is oriented around Jupyter notebooks instead of Sage worksheets. So the default file type is Jupyter notebooks, the default kernel doesn't involve the extra heft of Sage, etc., and the domain name involves Jupyter instead of "sagemath". Some people are disuased from using SageMathCloud for Jupyter notebooks because of the "SageMath" name.

Dozens of web applications (including SageMathCloud) use the word "Jupyter" in various places. However, I was unsure about using "jupyter" in a domain name. I found this github issue and requested clarification 6 weeks ago. We've had some back and forth, but they recently made it clear that it would be at least a month until any decision would be considered, since they are too busy with other things. In the meantime, I rented jupytercloud.com, which has a nice ring to it, as the planet Jupiter has clouds. Yesterday, I made jupytercloud.com point to cloud.sagemath.com to see what it would "feel like" and Tim Clemans started experimenting with customizing the page based on the domain name that the client sees. I did not mention jupytercloud.com publicly anywhere, and there were no links to it.

Today I received this message:


    I'm writing this representing the Jupyter project leadership
    and steering council. It has recently come to the Jupyter
    Steering Council's attention that the domain jupytercloud.com
    points to SageMathCloud. Do you own that domain? If so,
    we ask that you take the domain name down immediately, as
    it uses the Jupyter name.

I of course immediately complied. It is well within their rights to dictate how their name is used, and I am obsessive about scrupulously doing everything I can to respect people's intellectual property; with Sage we have put huge amounts of effort into honoring both the letter and spirit of copyright statements on open source software.

I'm writing this because it's unclear to me what people really want, and I have no idea what to do here.

1. Do you want something built on the same technology as SageMathCloud, but much more focused on Jupyter notebooks?

2. Does the name of the site matter to you?

3. What model should the Jupyter project use for their trademark? Something like Python? like Git?Like Linux?  Like Firefox?  Like the email program PINE?  Something else entirely?

4. Should I be worried about using Jupyter at all anywhere? E.g., in this blog post? As the default notebook for the SageMath project?

I appreciate any feedback.

Hacker News Discussion

Friday, July 22, 2016

DataDog's pricing: don't make the same mistake I made

I stupidly made a mistake recently by choosing to use DataDog for monitoring the infrastructure for my startup (SageMathCloud).

I got bit by their pricing UI design that looks similar to many other sites, but is different in a way that caused me to spend far more money than I expected.

I'm writing this post so that you won't make the same mistake I did.  As a product, DataDog is of course a lot of hard work to create, and they can try to charge whatever they want. However, my problem is that what they are going to charge was confusing and misleading to me.

I wanted to see some nice web-based data about my new autoscaled Kubernetes cluster, so I looked around at options. DataDog looked like a new and awesomely-priced service for seeing live logging. And when I looked (not carefully enough) at the pricing, it looked like only $15/month to monitor a bunch of machines. I'm naive about the cost of cloud monitoring -- I've been using Stackdriver on Google cloud platform for years, which is completely free (for now, though that will change), and I've also used self hosted open solutions, and some quite nice solutions I've written myself. So my expectations were way out of whack.

Ever busy, I signed up for the "$15/month plan":

One of the people on my team spent a little time and installed datadog on all the VM's in our cluster, and also made DataDog automatically start running on any nodes in our Kubernetes cluster. That's a lot of machines.

Today I got the first monthly bill, which is for the month that just happened. The cost was $639.19 USD charged to my credit card. I was really confused for a while, wondering if I had bought a year subscription.

After a while I realized that the cost is per host! When I looked at the pricing page the first time, I had just saw in big letters "$15", and "$18 month-to-month" and "up to 500 hosts". I completely missed the "Per Host" line, because I was so naive that I didn't think the price could possibly be that high.

I tried immediately to delete my credit card and cancel my plan, but the "Remove Card" button is greyed out, and it says you can "modify your subscription by contacting us at success@datadoghq.com":

So I wrote to success@datadoghq.com:

Dear Datadog,

Everybody on my team was completely mislead by your
horrible pricing description.

Please cancel the subscription for wstein immediately
and remove my credit card from your system.

This is the first time I've wasted this much money
by being misled by a website in my life.

I'm also very unhappy that I can't delete my credit
card or cancel my subscription via your website.  It's
like one more stripe API call to remove the credit card
(I know -- I implemented this same feature for my site).

And they responded:

Thanks for reaching out. If you'd like to cancel your
Datadog subscription, you're able to do so by going into
the platform under 'Plan and Usage' and choose the option
downgrade to 'Lite', that will insure your credit card
will not be charged in the future. Please be sure to
reduce your host count down to the (5) allowed under
the 'Lite' plan - those are the maximum allowed for
the free plan.

Also, please note you'll be charged for the hosts
monitored through this month. Please take a look at
our billing FAQ.

They were right -- I was able to uninstall the daemons, downgrade to Lite, remove my card, etc. all through the website without manual intervention.

When people have been confused with billing for my site, I have apologized, immediately refunded their money, and opened a ticket to make the UI clearer.  DataDog didn't do any of that.

I wish DataDog would at least clearly state that when you use their service you are potentially on the hook for an arbitrarily large charge for any month. Yes, if they had made that clear, they wouldn't have had me as a customer, so they are not incentivized to do so.

A fool and their money are soon parted. I hope this post reduces the chances you'll be a fool like me.  If you chose to use DataDog, and their monitoring tools are very impressive, I hope you'll be aware of the cost.


On Hacker News somebody asked: "How could their pricing page be clearer? It says per host in fairly large letters underneath it. I'm asking because I will be designing a similar page soon (that's also billed per host) and I'd like to avoid the same mistakes."  My answer:

[EDIT: This pricing page by the top poster in this thread is way better than I suggest below -- https://www.serverdensity.com/pricing/]

1. VERY clearly state that when you sign up for the service, then you are on the hook for up to $18*500 = $9000 + tax in charges for any month. Even Google compute engine (and Amazon) don't create such a trap, and have a clear explicit quota increase process.
2. Instead of "HUGE $15" newline "(small light) per host", put "HUGE $18 per host" all on the same line. It would easily fit. I don't even know how the $15/host datadog discount could ever really work, given that the number of hosts might constantly change and there is no prepayment.
3. Inform users clearly in the UI at any time how much they are going to owe for that month (so far), rather than surprising them at the end. Again, Google Cloud Platform has a very clear running total in their billing section, and any time you create a new VM it gives the exact amount that VM will cost per month.
4. If one works with a team, 3 is especially important. The reason that I had monitors on 50+ machines is that another person working on the project, who never looked at pricing or anything, just thought -- he I'll just set this up everywhere. He had no idea there was a per-machine fee.

Wednesday, February 24, 2016

Elliptic curves: Magma versus Sage

Elliptic Curves

Elliptic curves are certain types of nonsingular plane cubic curves, e.g., y^2 = x^3 + ax +b, which are central to both number theory and cryptography (e.g., they are used to compute the hash in bitcoin).

Magma and Sage

If you want to do a wide range of explicit computations with elliptic curves, for research purposes, you will very likely use SageMath or Magma. If you're really serious, you'll use both.

Both Sage and Magma are far ahead of all other software (e.g., Mathematica, Maple and Matlab) for elliptic curves.

A Little History

When I started contributing to Magma in 1999, I remember that Magma was way, way behind Pari. I remember having lunch with John Cannon (founder of Magma), and telling him I would no longer have to use Pari if only Magma would have dramatically faster code for computing point counts on elliptic curves.

A few years later, John wisely hired Mark Watkins to work fulltime on Magma, and Mark has been working there for over a decade. Mark is definitely one of the top people in the world at implementing (and using) computational number theory algorithms, and he's ensured that Magma can do a lot. Some of that "do a lot" means catching up with (and surpassing!) what was in Pari and Sage for a long time (e.g., point counting, p-adic L-functions, etc.)

However, in addition, many people have visited Sydney and added extremely deep functionality for doing higher descents to Magma, which is not available in any open source software. Search for Magma in this paper to see how, even today, there seems to be no open source way to compute the rank of the curve y2 = x3 + 169304x + 25788938.  (The rank is 0.)

Two Codebases

There are several elliptic curves algorithms available only in Magma (e.g., higher descents) ... and some available only in Sage (L-function rank bounds, some overconvergent modular symbols, zeros of L-functions, images of Galois representations). I could be wrong about functionality not being in Magma, since almost anything can get implemented in a year...

The code bases are almost completely separate, which is a very good thing. Any time something gets implemented in one, it gets (or should get) tested via a big run on elliptic curves up to some bound in the other. This sometimes results in bugs being found. I remember refereeing the "integral points" code in Sage by running it against all curves up to some bound and comparing to what Magma output, and getting many discrepancies, which showed that there were bugs in both Sage and Magma.
Thus we would be way better off if Sage could do everything Magma does (and vice versa).

Tuesday, February 23, 2016

"If you were new faculty, would you start something like SageMathCloud sooner?"

I was recently asked by a young academic: "If you were a new faculty member again, would you start something like SageMathCloud sooner or simply leave for industry?" The academic goes on to say "I am increasingly frustrated by continual evidence that it is more valuable to publish a litany of computational papers with no source code than to do the thankless task of developing a niche open source library; deep mathematical software is not appreciated by either mathematicians or the public."

I wanted to answer that "things have gotten better" since back in 2000 when I started as an academic who does computation. Unfortunately, I think they have gotten worse. I do not understand why. In fact, this evening I just received the most recent in a long string of rejections by the NSF.

Regarding a company versus taking a job in industry, for me personally there is no point in starting a company unless you have a goal that can only be accomplished via a company, since building a business from scratch is extremely hard and has little to do with math or research. I do have such a goal: "create a viable open source alternative to Mathematica, etc...". I was very clearly told by Michael Monagan (co-founder of Maplesoft) in 2006 that this goal could not be accomplished in academia, and I spent the last 10 years trying to prove him wrong.

On the other hand, leaving for a job in industry means that your focus will switch from "pure" research to solving concrete problems that make products better for customers. That said, many of the mathematicians who work on open source math software do so because they care so much about making the experience of using math software much better for the math community. What often drives Sage developers is exactly the sort of passionate care for "consumer focus" and products that also makes one successful in industry. I'm sure you know exactly what I mean, since it probably partly motivates your work. It is sad that the math community turns its back on such people. If the community were to systematically embrace them, instead of losing all these $300K+/year engineers to mathematics entirely -- which is exactly what we do constantly -- the experience of doing mathematics could be massively improved into the future. But that is not what the community has chosen to do. We are shooting ourselves in the foot.

Now that I have seen how academia works from the inside over 15 years I'm starting to understand a little why these things change very slowly, if ever. In the mathematics department I'm at, there are a small handful of research areas in pure math, and due to how hiring works (voting system, culture, etc.) we have spent the last 10 years hiring in those areas little by little (to replace people who die/retire/leave). I imagine most mathematics departments are very similar. "Open source software" is not one of those traditional areas. Nobody will win a Fields Medal in it.

Overall, the mathematical community does not value open source mathematical software in proportion to its value, and doesn't understand its importance to mathematical research and education. I would like to say that things have got a lot better over the last decade, but I don't think they have. My personal experience is that much of the "next generation" of mathematicians who would have changed how the math community approaches open source software are now in industry, or soon will be, and hence they have no impact on academic mathematical culture. Every one of my Ph.D. students are now at Google/Facebook/etc.

We as a community overall would be better off if, when considering how we build departments, we put "mathematical software writers" on an equal footing with "algebraic geometers". We should systematically consider quality open source software contributions on a potentially equal footing with publications in journals.

To answer the original question, YES, knowing what I know now, I really wish I had started something like SageMathCloud sooner. In fact, here's the previously private discussion from eight years ago when I almost did.


- There is a community generated followup ...

Wednesday, February 10, 2016

Open source is now ready to compete with Mathematica for use in the classroom

When I think about what makes SageMath different, one of the most fundamental things is that it was created by people who use it every day.  It was created by people doing research math, by people teaching math at universities, and by computer programmers and engineers using it for research.  It was created by people who really understand computational problems because we live them.  We understand the needs of math research, teaching courses, and managing an open source project that users can contribute to and customize to work for their own unique needs.

The tools we were using, like Mathematica, are clunky, very expensive, and just don't do everything we need.  And worst of all, they are closed source software, meaning that you can't even see how they work, and can't modify them to do what you really need.  For teaching math, professors get bogged down scheduling computer labs and arranging for their students to buy and install expensive software.

So I started SageMath as an open source project at Harvard in 2004, to solve the problem that other math software is expensive, closed source, and limited in functionality, and to create a powerful tool for the students in my classes.  It wasn't a project that was intended initially as something to be used by hundred of thousands of people.  But as I got into the project and as more professors and students started contributing to the project, I could clearly see that these weren't just problems that pissed me off, they were problems that made everyone angry.

The scope of SageMath rapidly expanded.  Our mission evolved to create a free open source serious competitor to Mathematica and similar closed software that the mathematics community was collective spending hundreds of millions of dollars on every year. After a decade of work by over 500 contributors, we made huge progress.

But installing SageMath was more difficult than ever.  It was at that point that I decided I needed to do something so that this groundbreaking software that people desperately needed could be shared with the world.

So I created SageMathCloud, which is an extremely powerful web-based collaborative way for people to easily use SageMath and other open source software such as LaTeX, R, and Jupyter notebooks easily in their teaching  and research.   I created SageMathCloud based on nearly two decades of experience using math software in the classroom and online, at Harvard, UC San Diego, and University of Washington.

SageMathCloud is commercial grade, hosted in Google's cloud, and very large classes are using it heavily right now.  It solves the installation problem by avoiding it altogether.  It is entirely open source.

Open source is now ready to directly compete with Mathematica for use in the classroom.  They told us we could never make something good enough for mass adoption, but we have made something even better.  For the first time, we're making it possible for you to easily use Python and R in your teaching instead of Mathematica; these are industry standard mainstream open source programming languages with strong support from Google, Microsoft and other industry leaders.   For the first time, we're making it possible for you to collaborate in real time and manage your course online using the same cutting edge software used by elite mathematicians at the best universities in the world.

A huge community in academia and in industry are all working together to make open source math software better at a breathtaking pace, and the traditional closed development model just can't keep up.

Friday, January 15, 2016

Thinking of using SageMathCloud in a college course?

SageMathCloud course subscriptions

"We are  college instructors of the calculus sequence and ODE’s.  If the college were to purchase one of the upgrades for us as we use Sage with our students, who gets the benefits of the upgrade?  Is is the individual students that are in an instructor’s Sage classroom or is it the  collaborators on an instructor’s project?"

If you were to purchase just the $7/month plan and apply the upgrades to *one* single project, then all collaborators on that one project would benefit from those upgrades while using that project.

If you were to purchase a course plan for say $399/semester, then you could apply the upgrades (network access and members only hosting) to 70 projects that you might create for a course.   When you create a course by clicking +New, then "Manage a Course", then add students, each student has their own project created automatically.  All instructors (anybody who is a collaborator on the project where you clicked "Manage a course") is also added to the student's project. In course settings you can easily apply the upgrades you purchase to all projects in the course.  

Also I'm currently working on a new feature where instructors may choose to require all students in their course to pay for the upgrade themselves.  There's a one time $9/course fee paid by the student and that's it.  At some colleges (in some places) this is ideal, and at other places it's not an option at all.   I anticipate releasing this very soon.

Getting started with SageMathCloud courses

You can fully use the SMC course functionality without paying anything in order to get familiar with it and test it out.  The main benefit of paying is that you get network access and all projects get moved to members only servers, which are much more robust; also, we greatly prioritize support for paying customers.   

This blog post is an overview of using SMC courses:


This has some screenshots and the second half is about courses:


Here are some video tutorials made by an instructor that used SMC with a large class in Iceland recently:


Note that the above videos show the basics of courses, then talk specifically about automated grading of Jupyter notebooks.  That might not be at all what you want to do -- many math courses use Sage worksheets, and probably don't automate the grading yet.

Regarding using Sage itself for teaching your courses, check out the free pdf book to "Sage for Undergraduates" here, which the American Mathematical Society just published (there is also a very nice print version for about $23):


Friday, January 8, 2016

Mathematics Graduate School: preparation for non-academic employment

This is about my personal experience as a mathematics professor whose students all have non-academic jobs that they love. This is in preparation for a panel at the Joint Mathematics Meetings in Seattle.

My students and industry
My graduated Ph.D. students:
  • 3 at Google
  • 1 at Facebook
  • 1 at CCR
My graduating student (Hao Chen):
  • Applying for many postdocs
  • But just did summer internship at Microsoft Research with Kristin. (I’ve had four students do summer internships with Kristin)
All my students:
  • Have done a lot of Software development, maybe having little to do with math, e.g., “developing the Cython compiler”, “transition the entire Sage project to git”, etc.
  • Did a thesis squarely in number theory, with significant theoretical content.
  • Guilt (or guilty pleasure?) spending time on some programming tasks instead of doing what they are “supposed” to do as math grad students.

Me: academia and industry

  • Math Ph.D. from Berkeley in 2000; many students of my advisor (Lenstra) went to work at CCR after graduating…
  • Academia: I’m a tenured math professor (since 2005) – number theory.
  • Industry: I founded a Delaware C Corp (SageMath, Inc.) one year ago to “commercialize Sage” due to VERY intense frustration trying to get grant funding for Sage development. Things have got so bad, with so many painful stupid missed opportunities over so many years, that I’ve given up on academia as a place to build Sage.
Reality check: Academia values basic research, not products. Industry builds concrete valuable products. Not understanding this is a recipe for pain (at least it has been for me).

Advice for students from students

Robert Miller (Google)

My student Robert Miller’s post on Facebook yesterday: “I LOVE MY JOB”. Why: “Today I gave the first talk in a seminar I organized to discuss this result: ‘Graph Isomorphism in Quasipolynomial Time’. Dozens of people showed up, it was awesome!”
Background: When he was my number theory student, working on elliptic curves, he gave a talk about graph theory in Sage at a Sage Days (at IPAM). His interest there was mainly in helping an undergrad (Emily Kirkman) with a Sage dev project I hired her to work on. David Harvey asked: “what’s so hard about implementing graph isomorphism”, and Robert wanted to find out, so he spent months doing a full implementation of Brendan McKay’s algorithm (the only other one). This had absolutely nothing to do with his Ph.D. thesis work on the Birch and Swinnerton-Dyer conjecture, but I was very supportive.

Craig Citro (Google)

Craig Citro did a Ph.D. in number theory (with Hida), but also worked on Sage aLOT as a grad student and postdoc. He’s done a lot of hiring at Google. He says: “My main piece of advice to potential google applicants is ‘start writing as much code as you can, right now.’ Find out whether you’d actually enjoyworking for a company like Google, where a large chunk of your job may be coding in front of a screen. I’ve had several friends from math discover (the hard way) that they don’t really enjoy full-time programming (any more than they enjoy full-time teaching?).”
“Start throwing things on github now. Potential interviewers are going to check out your github profile; having some cool stuff at the top is great, but seeing a regular stream of commits is also a useful signal.”

Robert Bradshaw (Google)

“A lot of mathematicians are good at (and enjoy) programming. Many of them aren’t (and don’t). Find out. Being involved in Sage is significantly more than just having taken a suite of programming courses or hacking personal scripts on your own: code reviews, managing bugs, testing, large-scale design, working with others’ code, seeing projects through to completion, and collaborating with others, local and remote, on large, technical projects are all important. It demonstrates your passion.”

Rado Kirov (Google)

Robert Bradshaw said it before me, but I have to repeat. Large scale software development requires exposure to a lot of tooling and process beyond just writing code - version control, code reviews, bug tracking, code maintenance, release process, coordinating with collaborators. Contributing to an active open-source project with a large number of contributors like Sage, is a great way to experience all that and see if you would like to make it your profession. A lot of mathematicians write clever code for their research, but if less than 10 people see it and use it, it is not a realistic representation of what working as a software engineer feels like. 

The software industry is in large demand of developers and hiring straight from academia is very common. Before I got hired by Google, the only software development experience on my resume was the Sage graph editor. Along with solid understanding of algorithms and data structures that was enough to get in."

David Moulton (Google)

“Google hires mathematicians now as quantitative analysts = data engineers. Google is very flexible for a tech company about the backgrounds of its employees. We have a long-standing reading group on category theory, and we’re about to start one on Babai’s recent quasi- polynomial-time algorithm for graph isomorphism. And we have a math discussion group with lots of interesting math on it.”

My advice for math professors

Obviously, encourage your students to get involved in open source projects like Sage, even if it appears to be a waste of time or distraction from their thesis work (this will likely feel very counterintuitive you’ll hate it).
At Univ of Washington, a few years ago I taught a graduate-level course on Sage development. The department then refused to run it again as a grad course, which was frankly very frustrating to me. This is exactly the wrong thing to do if you want to increase the options of your Ph.D. students for industry jobs. Maybe quit trying to train our students to be only math professors, and instead give them a much wider range of options.