Some new insight into Galaxy


#1

Meteor just posted a new position that may shed some light as to what may be in store for Galaxy:

https://www.meteor.com/jobs/cloud-systems-engineering

Interesting read - it sounds consistent to what Arunoda was saying about Kubernetes and Docker, and the automatic CDN looks interesting too!


Galaxy - Platform-as-a-service-as-a-service
#2

Wow! Let’s go through it…

Design an ingress tier for modern, stateful applications that is capable of supporting millions of simultaneous connections.
Add AWS EC2 support to Kubernetes.

Ergo AWS-hosted PaaS, as you suggested…

Build a scalable, automated, containerized MongoDB hosting service capable of running thousands of separate databases without ever losing data.

Yay! They are hosting (and NEVAR LOSING) the data, as I suggested would be essential to the value proposition.

Design and implement a responsive and stable autoscaling mechanism for apps with long-lived websocket connections.

Wow…on this point they make plain they are looking for a genius. This line may inspire the top candidate and scare off the more typical. Though I don’t know so much about what’s involved.

Make an API for x509 certificate management that is both secure and understandable to users.

Turn-key PaaS. Cradle-to-grave. As you suggested.

Figure out what metrics are important to measure the performance of a real-time application and how to collect them.

Hey…we’ve already got this…

Implement automatic CDN support so that developers can improve application load-times with ease.

As you prophesied, meteor deploy and everything just works…

Design and implement a streaming log collection facility that reliably collects high volumes of log data without impacting application performance.

Here I’m wondering…does this have buy/build right? Can’t Papertrail or something be adapted pleasantly?

Anyway wow that’s a great spec and also wow they haven’t hired the maniac who will pull it off. So ETA 2016 or so after a long shake-out I’d suppose.

Thanks for the eagle eye @msavin


#3

Epic explanation @spicemix. Sounds like they are trying to lure in @arunoda :grin:

I hope MDG at least fixes the oplog issue until they find someone.


#4

They forgot to add “PaaS based space satellite deployment. NASA training included.”


#5

You’ve just beaten me to a speculative “Breaking news! MDG acquires meteorhacks!” post on meteor blog LOL :smile:


#6

@serkandurusoy Tomorrow is the day for it :slight_smile:


#7

I woke up with an idea as to how they’re going to do all of this. I would figure it uses a 2-tier architecture: a static tier and a stateful Meteor tier.

Static Tier

  • Redirects stateful requests to Meteor tier, load-balanced
  • Plays ping-pong to keep long-lived web socket connections open “for free” (scaling down)
  • Serves as firewall, anti-D/Dos
  • Handles TLS/Certificates
  • Redirects static content requests to CDN
  • Includes static “spiderable” sitemaps/content for robots
  • Edge deployed across geographies
  • Not written in Node…could be HAProxy, NginX, straight C/++

Meteor (Stateful) Tier

  • Meteor Node instances clustered around Mongo replica sets
  • Meteor & Mongo co-located in data centers for performance
  • Transparent sharding of Meteor/Mongo clusters by geography and user account
  • Node fibers spun up only for stateful requests, thus auto-scaling at fiber/thread granularity
  • Meteor apps no longer need to worry about everything static tier handles
  • Usage metered & billed beyond free “1 Meteorite” CPU share
  • Failover provisions at each sub-tier
  • Integrated logging with Static tier

So basically, the static tier takes care of security and the web socket connections, letting the apps people write be quite naive of all that and still work snappily. And that’s how they do the auto-scaling and billing even though the web sockets stay open when the app is spun down. Since MDG has control over the Meteor app format, they can make the whole thing just a matter of meteor deploy and some configuration pages on the web. Oh and a credit card! :wink:

As for Arunoda, Meteor already has one of those thankfully, cranking away on their own. So while I’m certainly not adverse to their bringing him into the fold, why have just one Arunoda if, for a limited amount of money, you can have two? I do think this is going to be so much work they will license Kadira from him rather than re-building their own version. In fact this may be at least the work of two if not three or four full-time staff to make this all spin like a top. In addition to the staff they already have filling out the platform itself.

Anyway it’s all moving in the right direction and with the right people will be quite amazing!


#8

Today is a good day for such an announcement!


#9

Great insight…i just wish the MDG were further advanced in their galaxy delivery (or perhaps they are and they are just adding to their team).

My take on the big picture…galaxy is likely to take a while…in the interim, @arunoda is developing his own galaxy equivalent - a paid PaaS offering that is based on his open source initiatives (cluster, kadira, mup etc). Maybe its just wishful thinking, but if so, i’m all for it. The Meteor ecosystem will benefit from the competition and developers will have greater choice as to where we deploy our apps. Such a case would be win win imo


#10

Yes there are several scenarios where Arunoda makes more money and Meteor is a lot more successful as a whole if he remains independent of MDG. There are also scenarios in which the opposite is true. I hope the best scenario happens for everyone!

Though as a practical matter, if he gets sucked into Galaxy we may have a gap where there aren’t good production deployment options until Galaxy is completed. I would urge others to fill that gap if so, because not only will Galaxy take a while (I estimate at least a year given this job listing) getting here, it may not be the best fit for all deployments when it does. And the deployments it doesn’t fit are likely to be high-budget ones. So it will remain lucrative to provide a 3rd party alternative both now and for a long time.


#11

MDG seems to be disadvantaged because deployment infrastructure is a commodity and it may not be their core competence. I think what they could provide what no one else can is the assurance that their solution is the best, or at least best supported.

Galaxy is positioned as an enterprise product - I anticipate it will be expensive and cater to corporate clients who will pay extra for the sense of security.

If there’s enough of a market I’d bet we’ll have open source alternative too. Arunoda is half way there. It may also be better suited for people who want more control and independence.


#12

Does it make sense for MDG to target Enterprise development? At some level, “Enterprise” will buy almost any development tool to some extent, but Meteor could be a bit more prepared for their needs:

  • There isn’t even SQL access yet
  • No training story
  • No established consultancy
  • No integration story with many EIS services
  • No channel presence
  • No high-profile Enterprise deployments

I worry that this would be losing by playing things too safe. I think instead MDG could successfully target “apps for startups” given that startups don’t have any of those integration needs or assistance norms, and MDG is a YC startup with many high-profile early-stage VCs on board. If they can network a few of those fellow startups to build atop Meteor, they become critical to those investments, and become invincible as a result.

I was coding in Scala a month ago, and while Typesafe is officially targeting Enterprise, and has far more of an Enterprise story in place, the thing that immortalizes them is Twitter and Linked In being built on Scala. That’s the way to win in tools today as an independent.


#13

I agree with @spicemix here. Plus, entrprise bears the unspoken connotation of “hard and expensive to implement” which Meteor isn’t. It is just too easy, therefore probably to be perceived almost a toy of a technology to find its place in the enterprise.

What’s ironic is scaling, Meteor’s only current gray area, is not actually that big an issue in the enterprise world, where we speak 10s of thousands of users at most.

I believe that MDG’s strategic move to embrace the windows world (which now increases Meteor’s span about 100-fold) will open up huge doors. People who have previously been afraid of programming because a mac/linux computer comes hard to get used to will give it a try. When they see how easy it is to develop a small app for their friends, family, small business, local karaoke group, backed by all the code-year kind of hype, they’ll be happy to spend away a couple of bucks on hosting their app with a single mouse click.

And I think that’s where the money lies. And I think its good.


#14

And there you go :slight_smile:


#15

I agree with @spicemix.
From my experience enterprise customers want:

  • Paid support, consulting and training
  • Proven performance in other enterprises (of course this is a bit chicken-egg).
  • Run on-premise/private cloud
  • If ran in the cloud, a lot of enterprises will prefer Azure over Amazon or Google.

#16

It could be worth to point out Meteor seems to have a close relationship with IBM, and a16z, their investor, has all the enterprise ties in the world.


#17

Except in Asia, most companies and public sectors are using Microsoft products and Sharepoint for websites. Just thought on convince them to choose Meteor is another challenges.


#18

IBM is a really interesting and out-of-place story.

The old IBM of the mainframe age was the dominant technology within the enterprise but interestingly enough, they missed out on the windows era transformation so Microsoft took the lion’s share from them.

At one point, they decided to go all out in with a new strategy to embrace the services model. They even acquired Price Waterhouse Coopers, a company whose largest revenue share came (back then) from auditing services.

IBM, now having ditched most of its hardware business, is concentrating on acqiring new businesses who are set on services as their primary offering and they also invest in interesting products that complement those services.

So, yeah, Meteor can be an interest for them, especially since they have very keen interest on the internet of things model and Meteor can be a game changer there.

But all in all, I still see Meteor gain more revenue and cover more base with Microsoft tech and services who also finally seems to have embraced the reality of services and open cloud.


#19

And now, it’s beta time!


#20

Just signed up, however I don’t think I’ll be contacted anytime soon:

Please note that we will initially focus on teams with critical applications in production, adding users as we fine tune.

Plus, my contact number is an Australian mobile :stuck_out_tongue:

But one can only hope :smile: