R&D for L&D

R&D for L&DBen Betts & James Mullaney

R&D for L&D

R&D for L&D

Scaling to meet demands

R&D for L&D

R&D for L&D

- Cloud based hardware (CentOS, Ubuntu etc)

- Load balancers (with SSL offloaded)

- Scale horizontally

- Nginx or Apache?

- PHP (FPM), Node or both!

Application Stack

R&D for L&D

Application Throughput

~ 70 ms~ 120 ms~ 600 ms~ 5,000 ms

1 =10 =

100 =1000

=

R&D for L&D

Database Stack

- ~9 servers for single sharded cluster

- 3 Replicas (Storage)- 3 Config (Metadata)- 3 MongoS (Routing) - Fully redundant

- Pick your shard key wisely…

- WiredTiger vs MMap

R&D for L&D

Database Throughput

R&D for L&D

Mongo Storage Engines

(with 2KB statement size)

3-4X compression through WiredTiger

R&D for L&D

Making use of scale

R&D for L&D

Our Data isn’t really BIG yet

R&D for L&D

Free MOOC starting AGAIN soon…Learn xAPI

www.curatr3.com