Riak usage at Apollo Group

Silicon valley riak meetup talk presentation

Transcript

1. Sajithkumar Kizhakkiniyil – Architect Riak @ Apollo Group
2. 2 § About Apollo Group § About Apollo Learning Platform § Why Riak? § How Riak is used @ Apollo § The experience § Q&A Agenda
3. 3 § Apollo Group is the parent company of University of Phoenix § Apollo Group owns many other universities besides UoP § Pioneer in online education – 340K Current students – 1600 Online courses § Apollo Education Service (AES) – To help higher learning institutions produce the most effective and efficient educational delivery systems possible. About Apollo Group
4. 4 § A highly Scalable LMS – Designed to be multi-tenant – Heavy and rich user interaction § Architecture – Large number of services – GWT based browser based application – Mobile client – No SQL and Relational and Graph stores – Riak, Mongodb, RDS, Marklogic About Apollo Learning Platform
5. 5 § Deployment – SaaS based delivery – Currently hosted in EC2 – Heavy integration with systems hosted in private Data Center – Released to UoP in August 2012 About Apollo Learning Platform (Cont…)
6. 6 § Why No SQL – Scalable – Access patterns § Why Riak – Evaluated Cassandra and Riak – Simple data model (key-value based) – Easy to add nodes to a cluster. – Linear scalability. Performance and storage scale as more nodes are added to the cluster – Full support for vector clocks allows conflicts to be resolved (although, the application has it share of responsibilities) – Easy to manage – Master less – CAP – Support Why Riak?
7. 7 § Configuration – 6 clusters across development and production – Level db – Largest 15 nodes, smallest 6 – 1024 ring size – All in EC2 – Started with 0.96 and now on Riak 1.14 – Replication and backup § Data – Domain object storage – Application metadata How Riak is Used @ Apollo
8. 8 § Access Pattern – Mostly keyed access § Not using – M/R – Used once backed off because of performance – 2i – Used once backed off because of performance – Key filter – No links How Riak is Used @ Apollo (cont…)
9. 9 §  Value added by Apollo –  Riak PB Driver • We have our implementation of PB. Early stages the PB driver did not have needed features • Added connection pooling • Connection pruning to shied EC2 socket behavior • Multiple cluster support • Advanced logging • Batch operations –  Distributed Cache • Service on top of Riak memory backend • Easy to implement and support • Have reasonably good performance –  Riak stats collection • List keys and List buckets are not usable at production • POST commit hook to collect data created • Application like list key and key can be build on top of it How Riak is Used @ Apollo (cont…)
10. 10 § Hosting and operation – Initial challenges • Knowledge gap • Configure and reconfigure ring sizes • Lack of monitoring • Sibling explosion – Now • Smooth operation in production • Help from Basho • Setting up Mutli-AZ replication • Self-sufficient • Still lack better monitoring The experience (cont…)
11. 11 § Support – Excellent from the beginning • Always there to help • Help us in production migration • Prompt response • Helped in developing post commit hook The experience (cont…)
12. 12 § Development – Initial challenges • Use of features • M/R , 2i and list keys used but later forced to remove • Developer training – What to use what not to use • Sibling explosion crashed production – Now • Normalized • Uses only for keyed access • No M/R, 2i etc • Experts review developer code to make sure it is safe • Started using more advanced customization using erlang The experience
13. 13 § Sajithkumar Kizhakkiniyil (skizhakk@apollogrp.edu) Contact
14. 14 Q&A