Jive Software stores over 300 million new items a day in HBase for a period of 30 days. Data growth contributed to severely stressed servers and reduced performance. We alleviated this stress by switching from a random row key to a composite salted time based key. We now have significantly reduced disk IOPS,reduced data churn, and increased retrieval performance by up to 500 times. This session will discuss how we monitored our slow HBase cluster, our row key designs and HBase schema optimizations, and writing custom HBase JRuby shell commands to support non-trivial row keys.