SageMath Cloud backend architecture talk at Sage Days 56

smcarch56

# SageMath Cloud backend architecture talk at Sage Days 56

## Guiding principles

 - A place for everybody to use *all* math-related software easily, especially (but not only!) open source

 - High availability: automatically survive failure of any proper subset of datacenters

 - Make it very hard to permanently lose work:
     - everything has synchronization
     - you can kill/close your browser at pretty much any moment without losing state (sharp contrast to ipython's design)

## Architecture

  Users <--...

  - Web Browser <-(todo: route 53)-> Stunnel: ssl (elliptic curve)
  - Stunnel <--> Haproxy:
         - websocket (or many old-fashioned fallbacks via Sock.js)
         - proxy servers
  - Haproxy <--> Hubs  (load balanced)
         - servers written in Node.js; several on each of 20+ VM's
  - Hub <--> Local Hub (one for each project)
         - simpler server written in node.js
         - 23 VM's run projects right now
  - Local Hub <--> Sage Server (one for each project)
  - Local Hub <--> IPython Servers (one for each directory of each project)
  - Local Hub <--> TTY Console server (node.js -- one for each project)

    ---> sage, consoles, files, ipython, etc.

- Database:

  - Cassandra (version 1.2.9):
     - distributed peer to peer database
     - no single points of failure
     - multi-datacenter aware
     - written in Java

- Distributed project storage:

  - Project is stored on computer where it runs, which minizes latency, while allowing for a slow large distributed encrypted network.

  - ZFS
     - next generation filesystem... but ready for production use (around for over 10 years)
     - file system *and* local volume managment (easy to expand with more disks)
     - snapshots
     - replication
     - compression
     - deduplication
     - verification

  - Each project stored in 2 locations in each data center
     - where?  Consistent hashing

- Network:

  - Tinc
     - the most mature general peer to peer VPN
     - no single points of failure
     - self-healing

- Synchronization:

  - I wrote (and BSD licensed!) a new implementation in CoffeeScript of Neil Fraser's Differential Synchronization algorithm: https://neil.fraser.name/writing/sync/
      - This is NOT "Operational Transforms"
  - All files, worksheets, etc. use this
  - Bolted it around IPython as well.


## Business

  - My motivation for SMC in May 2010:
      - meeting at Simons Foundation HQ on "Funding for Open Source Software in Math and Physics".
      - I want to generate more (and more sustainable) revenue to support Sage development.  I.e., I want to be able to hire *you*
        to work fulltime on Sage.  We can all want...

  - Discuss more later today...

## Functionality

  - Try it out and you'll see.  This could be for another talk tomorrow.