My notes: SXSWi: Scalability: covering your rear with a good backend

Scalability! Covering your rear with a good backend Full duplex cache Lld concatenated ssd forgot data Hbase over mysql (mysql for modeling)  mint runs mysql with sharding loves it. Design relational databases to scale horizontally Product first over architecture. If people aren't using your product what does the architecture matter. Want to write a book on scaling web services cookbook Hybridized approach dedicated database hardware Asynchronous processes Amazon Cloud front. Test images sent to asia Snmp management is still important to automate processes for integrity Qps ganglia cactus Always load test with real world data Simulate test data on small cluster When small build in cloud, but always design with scaling your data horizontally in mind, because you will grow out of it. hire diligent people (hard workers that will pick up skills quickly) Mint is heavily threaded architecture, all java based Have fun with what you do, but at the end of the day, does the product make sense? Imageshack nginx on 8 core, fast cgi, fast php Sidenote Nvidia datacenter - imageshack gets 2billion hits - hadoop and hive of petabytes of data, cuda cores  Grep for cuda. 438 cores. Map text into 3d framework and launch eternal map reduce job - bloom filter Never build for limitless scale from start, iterate often but plan for 12 months of configuration No one right architecture Time is the most important capital a startup has to manage Dev ops - integrating development and operations - mint is a fan  Shop kick - likes it Keep talent engaged  YouTube - hire good systems engineers after series a Cloud is not a substitute for hiring good engineers, depends on what infrastructure needs Netflix moved to the cloud because akamai said they would foot the bill - favorite tools - Hbase, nginx, memcached, haproxy replaced load balancer at yfrog - mint - good software engineer - shop kick - hers hadoop, scribes, zookeeper, thrift