Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Scaling SQL and NoSQL Databases in the Cloud


Published on

Database performance is the number-one cause of poor performance for scalable web applications, and the problem is magnified in cloud environments where I/O and bandwidth are generally slower and less predictable than in dedicated data centers. Database sharding is a highly effective method of removing the database scalability barrier by operating on top of proven RDBMS products such as MySQL and Postgres as well as the new NoSQL database platforms. In this session, you'll learn what it really takes to implement sharding, the role it plays in the effective end-to-end lifecycle management of your entire database environment, why it is crucial for ensuring reliability, and how to choose the best technology for a specific application. We'll also share a case study on a high-volume social networking application that demonstrates the effectiveness of database sharding for scaling cloud-based applications.

Published in: Technology

Scaling SQL and NoSQL Databases in the Cloud

  1. 1. Scaling your Database in the Cloud<br />Spring2011<br />Presented by:<br />Cory Isaacson, CEO<br />CodeFutures Corporation<br /><br />View the video of this presentation here!<br />
  2. 2. Introduction<br />Who I am<br />Cory Isaacson, CEO of CodeFutures<br />Providers of dbShards<br />Author of Software Pipelines<br />Partnerships:<br />Rightscale<br />The leading Cloud Management Platform<br />Leaders in database scalability, performance, and high-availability for the cloud…<br />…based on real-world experience with dozens of cloud-based applications<br />…social networking, gaming, data collection, mobile, analytics<br />Objective is to provide useful experience you can apply to scaling (and managing) your database tier…<br />…especially for high volume applications<br />
  3. 3. Challenges of cloud computing<br />Cloud provides highly attractive service environment<br />Flexible, scales with need (up or down)<br />No need for dedicated IT staff, fixed facility costs<br />Pay-as-you-go model<br />Cloud services occasionally fail<br />Partial network outages<br />Server failures…<br />…by their nature cloud servers are “transient”<br />Disk volume issues<br />Cloud-based resources are constrained<br />CPU<br />I/O Rates…<br />…the “Cloud I/O Barrier”<br />
  4. 4. Typical Application Architecture<br />
  5. 5. Scaling in the Cloud<br />Scaling Load Balancers is easy<br />Stateless routing to app server<br />Can add redundant Load Balancers if needed<br />DNS round-robin or intelligent routing for larger sites<br />If one goes down…<br />…failover to another<br />Scaling Application Servers is easy<br />Stateless<br />Sessions can easily transition to another server<br />Add or remove servers as need dictates<br />If one goes down…<br />…failover to another<br />
  6. 6. Scaling in the Cloud<br />Scaling the Database tier is hard<br />“Stateful” by definition (and necessity…)<br />Large, integrated data sets…<br />10s of GBs to TBs (or more…)<br />Difficult to move, reload<br />I/O dependent…<br />…adversely affected by cloud service failures<br />…and slow cloud I/O<br />If one goes down…<br />…ouch!<br />Databases form the “last mile” of true application scalability<br />Initially simple optimization produces the best result<br />Implement a follow-on scalability strategy for long-term performance goals…<br />…plus a high-availability strategy is a must<br />
  7. 7. All CPUs wait at the same speed…<br />The Cloud I/O Barrier<br />
  8. 8. More Database scalability challenges<br />Databases have many other challenges that limit scalability<br />ACID compliance…<br />…especially Consistency<br />…user contention<br />Operational challenges<br />Failover<br />Planned, unplanned<br />Maintenance<br />Index rebuild<br />Restore<br />Space reclamation<br />Lifecycle<br />Reliable Backup/Restore<br />Monitoring<br />Application Updates<br />Management<br />
  9. 9. Database slowdown is not linear…<br />
  10. 10. Challenges apply to all types of databases<br />Traditional RDBMS (MySQL, Postgres, Oracle…)<br />I/O bound<br />Multi-user, lock contention<br />High-availability<br />Lifecycle management<br />NoSQLDatabases In-memory, Caching, Document…)<br />Reliability, High-availability<br />Limits of a single server<br />…and a single thread<br />Data dumps to disk<br />Lifecycle Management<br />No matter what the technology, big databases are hard to manage…<br />…elastic scaling is a real challenge<br />…degradation from growth in size and volume is a certainty<br />Application-specific database requirements add to the challenge<br />…database design is key…<br />…balance performance vs. convenience vs. data size <br />
  11. 11. The Laws of Databases<br />Law #1: Small Databases are fast…<br />Law #2: Big Databases are slow…<br />Law #3: Keep databases small<br />
  12. 12. What is the answer?<br />Database sharding is the only effective method for achieving scale, elasticity, reliability and easy management…<br />…regardless of your database technology<br />
  13. 13. What is Database Sharding?<br />“Horizontal partitioning is a database design principle whereby rows of a database table are held separately... Each partition forms part of a shard, which may in turn be located on a separate database server or physical location.” Wikipedia<br />
  14. 14. The key to Database Sharding…<br />
  15. 15. Database Sharding Architecture<br />
  16. 16. Database Sharding Architecture<br />
  17. 17. Database Sharding… the results<br />
  18. 18. Why does Database Sharding work?<br />Maximize CPU/Memory per database instance…<br />…as compared to database size<br />Reduce the size of index trees<br />Speeds writes dramatically<br />No contention between servers<br />Locking, disk, memory, CPU<br />Allows for intelligent parallel processing…<br />…Go Fish queries across shards<br />Keep CPUs busy and productive<br />
  19. 19. Breaking the Cloud I/O Barrier<br />
  20. 20. Database types<br />Traditional RDBMS<br />Proven<br />SQL language…<br />…although ORM can change that<br />Durable, transactional, dependable…<br />…perceived as inflexible<br />NoSQL<br />Typically in-memory…<br />…the real speed advantage<br />Various flavors…<br />Key/Value Store<br />Document database<br />Hybrid<br />Store complex documents using a key<br />In-memory and disk based<br />
  21. 21. Black box vs. Relational Sharding<br />Both utilize sharding on a data key…<br />…typically modulus on a value or consistent hash<br />Black box sharding is automatic…<br /> …attempts to evenly shard across all available servers<br />…no developer visibility or control<br />…can work acceptably for simple, non-indexed NoSQL data stores<br />…easily supports single-row/object results<br />Relational sharding is defined by the developer…<br />…selective sharding of large data<br />…explicit developer control and visibility<br />…tunable as the database grows and matures<br />…more efficient for result sets and searchable queries<br />…preserves data relationships<br />
  22. 22. How Database Sharding works for NoSQL<br />
  23. 23. Relational Sharding<br />Shard-Tree Root Table<br />Shard-Tree Child Tables<br />Global Tables<br />
  24. 24. How Relational Sharding works<br />
  25. 25. How Relational Sharding works<br />Shard key recognition in SQL<br />SELECT * FROM customerWHERE customer_id = 1234<br />INSERT INTO customer(customer_id, first_name, last_name, addr_line1,…)VALUES(2345, ‘John’, ‘Jones’, ‘123 B Street’,…)<br />UPDATE customerSET addr_line1 = ‘456 C Avenue’WHERE customer_id = 4567<br />
  26. 26. What about Cross-Shard result sets?<br />
  27. 27. Cross-shard result set example<br />Go Fish (no shard key)<br />SELECT country_id, count(*) FROM customerGROUP BY country_id<br />
  28. 28. More on Cross-Shard result sets…<br />Black Box approach requires “scatter gather” for multi-row/object result set…<br />…common with NoSQL engines<br />…forces use of denormalized lists<br />…must be maintained by developer/application code<br />…Map Reduce processing helps with this (non-realtime)<br />Relational sharding provides access to meaningful result sets of related data…<br />…aggregation, sort easier to perform<br />…logical search operations more natural<br />…related rows are stored together<br />
  29. 29. Make your Application Shard-Safe<br />For sharding to be efficient, every sharded table should include a shard key.<br />SELECT WHERE clause statements should include a shard key where possible so that the query goes directly to a single shard rather than being executed against multiple shards in parallel.<br />INSERT, UPDATE and DELETE WHERE CLAUSE statements must include a shard key so that the data is written to the correct shard.<br />Avoid cross-shard inner joins. For example, joining one customer to another is not feasible as both customers may be in different shards. This can be accomplished, but requires complex logic and can adversely affect performance.<br />
  30. 30. What about High-Availability?<br />Can you afford to take your databases offline:<br />…for scheduled maintenance? <br />…for unplanned failure? <br />…can you accept some lost transactions?<br />By definition Database Sharding adds failure points to the data tier<br />A proven High-Availability strategy is a must…<br />…system outages…especially in the cloud<br />…planned maintenance…a necessity for all database engines<br />
  31. 31. Traditional asynch replication is inadequate…<br /><ul><li>Replication “Lag” can result in lost transactions
  32. 32. No rapid failover for continuous operation
  33. 33. Maintenance difficult in production environments
  34. 34. Used by many NoSQL engines</li></li></ul><li>Database Sharding and High-Availability<br /><ul><li>Need a minimum of 2 active-active instances per shard…
  35. 35. …3 active-active instances for in-memory databases
  36. 36. Support for…
  37. 37. …fail-down
  38. 38. …failover
  39. 39. …maintenance
  40. 40. …automated backup/restore
  41. 41. …monitoring</li></li></ul><li>Database Sharding…elastic shards<br />
  42. 42. Database Sharding…elastic shards<br />Expand the number of shards…<br />…divide a single shard into N new shards<br />…or rebalance existing shards across N new shards<br />Contract the number of shards…<br />…consolidate N shards into a single shard<br />Due to large data sizes, this takes time…<br />…regardless of data architecture<br />Requires High-Availability to ensure no downtime…<br />…perform scaling on live replica in background<br />…fail back to active instances<br />
  43. 43. Database Sharding…the future<br />Ability to leverage proven database engines…<br />…SQL<br />…NoSQL<br />…Caching<br />Hybrid database infrastructures using the best of all technologies<br />In-memory and disk-based solutions co-existing<br />Allow developers to select the best database engine for a given set of application requirements…<br />…seamless context-switching within the application<br />…use the API of choice<br />Improved management…<br />…monitoring<br />…configuration “on-the-fly”<br />…dynamic elastic shards based on demand<br />
  44. 44. Database Sharding summary<br />Database Sharding is the only effective tool for scaling your database tier in the cloud…<br />…or anywhere<br />Use Database Sharding for any type of engine…<br />…SQL<br />…NoSQL<br />…Cache<br />Use the best engine for a given application requirement…<br />…avoid the “one-size-fits-all” trap<br />…each engine has its strengths to capitalize on<br />Ensure your High-Availability is proven and bulletproof…<br />…especially critical on the cloud<br />…must support failure and maintenance<br />
  45. 45. Questions/Answers<br />Cory Isaacson<br />CodeFutures Corporation<br /><br /><br />