Google Cloud Platform’s Pi API – With access to 750B digits of Pi | Hacker News Reader

Google Cloud Platform’s Pi API – With access to 750B digits of Pi

pi.delivery

214 pointsalpb87 comments

I'm calling this: after attending Google Cloud Next last week, and seeing the huge improvements to GCP in the last 18 months, I am supremely confident that AWS should be worried about Google as a legitimate competitor.

$GOOG is betting big on GCP, and they are executing at a very high level releasing iterative products and new features.

I'm recommending GCP for all my clients (shameless plug https://elasticbyte.net) if they don't have any infrastructure setup or provisioned already. While AWS offers a wider array of integrated services (they do have a large lead time), GCP has learned from all the mistakes of other providers. They started off with a superior project based structure, then per-minute billing, then sustained-use discounts, then Container Engine (GKE), then the Firebase acquisition (genius), now Functions and SQL Postgres support.

Most folks on HN are part of the startup crowd, and what that crowd misses is that while GCP is technically superior in every way, they have no idea how to sell to an enterprise.

Google's sales method is, "There's the docs, good luck". Enterprises don't like that. They like to be hand held, they like lots of humans taking them out for lunches and dinners.

So far Google has not shown any inclination that they know how to enterprise, and in fact it's fairly counter-cultural for them.

If they learn to sell to the enterprise, I agree with you.

But right now, almost all the money in cloud is with the enterprise, and right now Google has no idea how to address that market.

> So far Google has not shown any inclination that they know how to enterprise, and in fact it's fairly counter-cultural for them.

While traditionally true with Google Products, I disagree with Cloud. Did you attend Google Cloud Next? Trust me, they are focusing on enterprises. Their support has improved significantly and they are focusing on business cases (costs, management, audit, security).

I am certain the Google Cloud Next conference in a few years will look just as enterprise as AWS re:Invent.

I did attend briefly, and I got a run down from some enterprise folks. They said Google appears to be thinking about making an effort towards enterprise, but they aren't there yet.

Also, they don't have a sales force like Amazon. A bunch of non-engineer humans who do nothing but convince other people to buy their product. They are still trying to solve human problems with engineering, and sales is a tough nut to crack for a computer, even a really strong AI. :)

> Also, they don't have a sales force like Amazon. A bunch of non-engineer humans who do nothing but convince other people to buy their product.

Here's the catch, the "bunch of non-engineer humans" as you called them are not stupid though. All around these people are watching the landscape and market. They are aware that GCP has a superior pricing structure and can be up to 50% cheaper[1]. Therefore my theory is that GCP can continue to offer lower prices, greater governance, superior performance, and solid support, and their (sales+engineering) force can be in-bound driven instead of out-bound like AWS.

[1] - https://thehftguy.com/2016/11/18/google-cloud-is-50-cheaper-...

That's not how enterprise sales cycles work though. They don't call you, you call them. Even when they want the hot new thing, they shoot an email off and say, "please have a salesperson here to make a presentation on why we should switch to you".

And at the moment, AWS has them beat hands down on governance. It's one area where GCP still needs to play catch up, and it's really important for enterprises. I'm sure they'll get there, but they aren't there yet.

"The numbers given in this article do not account for any AWS reservation. However, they all account for Google sustained use discount (30% automatic discount on instances that ran for the entire month)."

I'll never understand why people keep referencing that 50% cheaper article. They take into account the 30% monthly discount from Google, but ignore the 47% discount I currently get from Amazon by reserving my instances for a year. Why? Well, because the author thinks "Reserved Instances are bullshit!".

Now, I agree reserved instances are not ideal, but if you're trying to write an article about pricing, it's absurd to exclude such a discount. It's obvious the author was just going for click-bait.

Why? RI's require upfront purchasing, sustained use doesn't. So the comparison is valid because it's comparing on-demand instances from both providers, and Google comes out cheaper.

Otherwise you can say anything has better pricing because you signed a big volume discount upfront, whether you're buying cloud VMs or tires.

AWS RIs don't require upfront purchasing (payment). They offer full-upfront; they offer half-upfront; they also offer no-upfront.

Source: https://aws.amazon.com/ec2/pricing/reserved-instances/pricin...

Yes, but it's still upfront purchasing. You don't have to pay immediately but you have to commit to the full term to get the discount, with an option to get more of a discount if you pay early too.

Google's sustained-use discounts have no commitments at all.

Google has released "Committed Use Discounts", an RI competitor, which appears to be ~35% cheaper than RIs on average, according to Rightscale [0].

I also wrote an article discussing why "Committed Use Discounts" is a much bigger deal than just a "price model" at [1].

It'd be great for "thefht" guy to do the comparison, but data's already out there.

(work at Google Cloud)

[0] http://www.rightscale.com/blog/cloud-cost-analysis/aws-reser...

[1] https://hackernoon.com/why-googles-answer-to-aws-reseved-ins...

That 47% discount is a discount on AWS on demand (EC2/RDS) prices - not a 47% discount relative to GCP sustained use pricing.

I just finished researching pricing on AWS/GCP for my team - and I found that generally AWS matches GCP pricing (sometimes a little higher, sometimes a little lower) if you can commit to 1 year.

From what I found through my research, the only way to achieve a significant reduction in price relative to GCP is if you can commit to 3 years on AWS.

However, I found it somewhat difficult to compare apples to apples - as it can be difficult to match CPU and RAM - I had to settle for close enough. There is a very big difference in network bandwidth allowances per type - with GCP being far more generous. Load balancing offerings differ greatly - with GCP seeming much more modern in design (HA, geo-load balancing, anycast static ip, etc.). My point here being that our analysis took into account price as well as other factors.

GCP has a sales force - one can debate their relative sophistication or effectiveness compared to Amazon, but Diane Greene has brought in a bunch of ex-VMware leadership to run the sales, presales, and consulting organizations, and they definitely understand enterprise sales.

Amazon just has some additional experience from doing it for longer.

At my last job I actually talked to a sales person from Google who tried really hard to get us to move to GCP.

Did they succeed?

Partly. They created a VPN between AWS and GCP and moved some data warehouse/analytics stuff to Google. This is after I left, so I don't know the details.

Did you even watch GCPNext? The entire event was enterprise-focused. They brought a half-dozen companies on stage to basically say the same thing over and over again: "We moved to Google cloud, its so much better, Google helped us every step of the way."

Its possible these companies were lying, but I think that conference should be evidence enough for any prospective customers that Google is ready.

> The entire event was enterprise-focused

Some very old-school enterprises are working with Google to onboard GCP in their organisations. A few of them even spoke at the event. I'm assuming they've done due diligence to ensure that Google will devote time and attention span to GCP and not treat it like a "side project".

Also, looking at this logically, whether Google offers GCP or not, you can't deny that that they'll continue to refine their internal infrastructure. Their internal platform is one of their crown jewels. It just makes sense to offer that to businesses willing to pay for it.

Yes. Sales is easy to staff up but technology is very hard to build. Google is going in the right order and can easily hire the enterprise handlers it needs to close large contracts.

I'm not saying Google can't do it, and I agree that they did it right (build the technology first). What I'm saying is that there isn't strong evidence that they are getting good at enterprise sales yet. If they can do it, then they will be very successful. I'm starting to see signs that they might be doing it right, but the jury is still out.

I did, yes. They are starting to show signs of learning how to sell to an enterprise. If they get it right, they will do well. I'm just saying the jury is still out.

>So far Google has not shown any inclination that they know how to enterprise, and in fact it's fairly counter-cultural for them.

They have over 3 million businesses that use G Suite and their other enterprise offerings. To imply that they no inclination on "how to enterprise" or that "it's fairly counter-cultural for them" is disingenuous to say the least.

>But right now, almost all the money in cloud is with the enterprise, and right now Google has no idea how to address that market.

That's amusing considering all of the enterprise customers that are using GCP. Perhaps you should watch their cloud conference.

Selling GSuite is very different than selling compute. Besides being SaaS vs. infrastructure, the GSuite contracts are tiny compared to enterprise infrastructure. If a company with 100,000 employees paid list price for GSuite monthly, that would only be $1,000,000/mo, which is peanuts for an enterprise contract. Especially since they would probably get a huge bulk discount.

What I mean by counter-cultural is that at Google, they try to solve problems with computers instead of humans. It's worked really well for them so far, but it won't work for selling huge enterprise contracts. Yes, they have spun up a small support and sales team, but they will need to grow it a lot to support lots of large enterprises. Also, in companies that cater to enterprises, the sales team drives the engineering roadmap, or it is at least a collaboration. From talking to folks in Google, it seems that they don't really allow the sales team any input in to product cycles or technology.

And I attended their cloud conference and I'm very aware of their customer base vs. AWS. Their customers would all be considered fairly small to AWS, with the exception of a few big players.

Edit: An additional anecdote, I saw firsthand how they support an enterprise install of thousands of GSuite users. For the most part, it was "there's the docs, good luck". Sometimes they would assign an engineer, who would never come on site, to tell us that what we want is impossible.

I'm guessing you didn't watch any of the videos of Google Next and haven't been paying attention to their moves for at least a year or two? Their new Customer Reliability Engineering team is no flipping joke. They helped scale out Pokemon Go (among other things).

https://cloudplatform.googleblog.com/2016/09/bringing-Pokemo...

Seems like an awful lot of handholding to me, so I suspect your argument is simply invalid.

I've been paying very close attention to what they are doing and attended Google Next. I know what they are doing. They seem to be doing the right thing, but the jury is still out. They still haven't proven they are good at the enterprise.

This is why they brought on Diane Greene who was the founder/CEO of VMWare. She clearly understands enterprise sales, has staffed up and merged departments appropriately and it's starting to show in everything from marketing to the support interactions.

> Most folks on HN are part of the startup crowd, and what that crowd misses is that while GCP is technically superior in every way, they have no idea how to sell to an enterprise.

They seem to be doing the build technology and sell to and partner with forms that specialize in enterprise sales thing, while building their own enterprise sales capacity.

Lots of players now dominant in enterprise and very big on direct enterprise sales (including Microsoft) got to that position largely by the same route, unseating predecessors who had much more expertise in direct enterprise sales.

This is a fair assessment. It will be interesting to see how it plays out now that they have a semi-complete enterprise offering.

They're well aware and have a plan for that as well.

There was an event recently at Google Munich where they offered a whole day of free Google Cloud training. I couldn't attend, so I can't comment on the quality.

It's cultural. Amazon's culture is all about "customer obsession". That's their thing. When they have breakthrough tech products (Kindle, Alexa, etc) it's because they try to focus on the customers needs (and actually get it right, unlike the fire phone).

AWS has an enterprise sales team, many of whom have strong engineering backgrounds, because that's what customer obsession looks like in the cloud world.

Whether you're on Amazon or Google, one thing is certain, we're going to have some amazing services and options available due to the competition between the two, and both of them fighting for market share.

I'm on AWS at the moment, and I'm in no hurry to jump ship, because I know Amazon is going to be forced to stay competitive with comparable services and pricing. In 3 or 5 years, if there's a clear winner, I'll make the switch, but for now, it's going to be fun just watching everything unfold.

I don't think switching will be the answer, but rather federation. We are currently deploying new services on Kubernetes and we can run isolated Kubernetes clusters on either GCP (not just GKE), or on AWS.

While we aren't to the point of using cluster federation for production yet, this is the next logical step. I see a, not to distant, future where we have cluster federation across on premise, GCP, and AWS clusters.

If, like us, your all in on the container Kool-Aid, then I believe the future looks a lot like policy based Kubernetes scheduling across federated clusters. Policy factors like geolocation, spot pricing, resource requirements, SLOs, etc... are all configureable to the app developer and transparently effected by pluggable Kubernetes schedulers.

Is it practical to replicate your data between clouds?

Legitimate competitor? With all of the data centers they have sprouting up this year and their superior networking technology and infrastructure I'll go on the record and say they'll be #1 in 5 years.

Completely agree they absolutely nailed networking and VPCs. GCP network attached disks are so much faster and easier to groc than EBS. The bigger the GCP disk, the faster it performs. If you want to maximize IOPS make a disk 750GB or larger. Compared that with EBS that has overhead with concepts like enhanced networking, EBS optimized, and don't even get me started on provisioned IOPS.

Provisioned IOPS is still there, but in AWS the standard disk is 'the bigger the disk, the faster it is' - 3IOPS/GB (theoretically), with max IOPS at 1000GB. 'Enhanced Networking' is on most server types now as well.

AWS pricing is definitely not made to be simple to understand, though...

Without functioning customer support all that is moot.

But they have support and it is actually awesome. What support plan are you on? Silver, Gold?

Have you even used the service?

Currently a customer. But thank you for asking.

Terrance from GCP Support.

If you have problems with our support team send me an email (tsg@google.com) and I will look into it.

So you know about this: https://cloud.google.com/support/

What did you mean by no functioning customer support?

So my take home here is that webscale makes https://github.com/philipl/pifs possible.

Now, if only there was a service to leftpad my strings...

Not sure if you'r being scarcastic, but... https://api.left-pad.io/?str=be%20careful%20what%20you%20wis...

That's great! I'd like to use this but sending my enterprise data to a third party is not very secure. Do they provide a docker image or ami?

Now, I just assume you're being sarcastic :). But, if you would like, I can build docker image that does the same thing for a price ;).

> This is just for fun. There is NO SLA. It might be turned off at any time!

Google service confirmed!

True.

loooollll

I'm a Google employee and you made me LOL hahaha

Is there anything we can learn from the transition between numbers? Any pattern?

Things I would do if I were a better JS programmer:

1. Figure out how many BPM I need to to have the music simulation generate digits at the same rate as the visual simulation.

2. Figure out how to press both start buttons at the same time.

it's actually here: https://pi.delivery/demo/

but we didn't have time to make it look nice :(

Awesome!! I knew I couldn't be the only one with that idea.

Is that a standard template you're using for the website?

not standard, but browsed through jekyll templates - this one fit our needs the most: https://github.com/CloudCannon/aviator-jekyll-template

I've always wondered if you take the digits from Pi, do they represent a pseudo random sequence or not?

It's speculated that Pi is a normal number[1]. If it is in fact normal, then I believe the answer is yes. However determining whether a number is normal or not is an unresolved problem according to Wolfram, so the answer is a solid maybe.

[1] http://mathworld.wolfram.com/NormalNumber.html

Pi is conjectured, but not proven to be a "normal number"[0] which is probably what you mean by a pseudorandom sequence.

[0] https://en.m.wikipedia.org/wiki/Normal_number

Just curious, what algorithms do you use for computing Pi? I have read about the Fabrice Bellard's job and algorithms about computing Pi, he uses relatively small CPU but his algorithms seem efficient.

The software they used is linked and has a some info about the internals here: http://www.numberworld.org/y-cruncher/algorithms.html

Pretty cool. I did look at my calendar quickly to see that it's not April 1. The "How we made this" part makes for a great parody of cloud computing.

> great parody of cloud computing

I don't get it.

All that stack and overhead to serve what is basically static data? Come on, if this was released ~2 weeks from now it would be funny...

It kind of reminds me when all bitcoin private keys were leaked: http://directory.io/

Edit: It's impressive, but still funny.

One of the great things about having all of this great tooling is it makes the easiest way to do something also the best way. They were able to quickly spin something up which, incidentally, could also massively scale should it need to.

Love that the piano works as a normal piano!

Listening to the essentially random sequence of notes, fancy recognizing Feynman's point

What was Feynman's point? I've got Pi/Listen going in the background, and it's quite pleasant, so I'm doubly interested.

https://en.wikipedia.org/wiki/Six_nines_in_pi

It's amusing to read the talk page on this article - there was lengthy investigation about whether this really had any Feynman connection (i.e. was the story about him apocryphal) and what the article should be called. I think the current title is fantastic - I would have struggled to name such a thing.

A sequence of many 9's in a row in the decimal expansion, which Feynman jokingly extended ... and 'showed' therefore pi was rational.

> I've got Pi/Listen going in the background, and it's quite pleasant That is probably due to the choice of musical scale more than due to Pi. I believe the output of randint(0,10) on the same musical scale would sound just as pleasant.

Nice job! You have a lot of AltCN in your SSL cert!

Here's the requisite cynical comment lamenting the death of "standards".

They could easily have implemented RFC 3091 [0], but instead they chose to create yet another proprietary API with vendor lock in, just as cloud service providers love to do.

[0]: https://tools.ietf.org/html/rfc3091

Mentioned above RFC specifies port number 314159 which is greater than possible max 65535. So once again unrealistic standards that does not work force people to invent something sane. /s

That is only a problem if you insist that Windows 3.1 is the best operating system in existence and if you are therefore unwilling to upgrade to Windows 95. The future is 32 bit, move on.

That first visualization using d3 is a classic example of a bad visualization isn't it? It's extremely difficult to get a sense of the difference in off-diagonal transition rates, and impossible to get an accurate sense of them or any sense at all of the on-diagonals. And yet, the transition rates could have been conveyed perfectly well by writing them in a 10 by 10 table! Maybe shading the cells of the table somehow to show the rate -- I'm no expert in data visualization -- just not the one chosen!

I think d3 is wonderful, but I worry that this sort of thing will make some people shake their heads and say "fancy JavaScript nonsense just causes people to make bad visualizations nowadays".

It's just supposed to look pretty, not really serve a function. We opened the API to the world so people who are way better at visualizations can build what they want!

I know it's Pi day, and the title uses the word for the greek letter, but my mind read the domain and immediately thought Google was offering an API for automated pie delivery.

Now I really want this.

I thought it was someone with a large array of raspberry pi computers for rent. It would be like an itty bitty AWS with adorable APIs at cute as a button prices.

There's a great April Fool's joke in there somewhere.

https://www.google.com/express/u/0/product/86239075729442885...

I was expecting Raspberry Pi

Google Cloud Pi Compute is coming out on April 1

Someone told me it was coming out last year April 1. I'm still waiting. Hope they do it this year.

Reply on news.ycombinator.com