When we introduced Non-Stop for Apache HBase, we explained how it would improve HBase reliability for critical applications. But Non-Stop for Apache HBase also uniquely improves HBase scalability and performance.
By making multiple active-active region servers, Non-Stop for Apache HBase alleviates some common HBase performance woes. First, clients are load balanced between several region servers for any particular region. By spreading the load among several region servers, the impact of problems like region ‘hot spots’ is alleviated.
So far so good, but you might be thinking that you could get the same benefit by using HBase read-HA. However, HBase read-HA is limited to read operations in a single data center. Non-Stop for Apache HBase lets you put region servers in several data centers, and any of them can handle write operations. That gives you a few nice benefits:
- Writes can be directed to any region server, reducing the chance that a single region server becomes a bottleneck due to hot spots or garbage collection.
- Applications at other data centers now have fast access to a ‘local’ region server.
Although the HBase community continues to try to improve HBase performance, there are some bottlenecks that just can’t be eliminated without active-active replication. No other solution lets you use several active region servers per region, and put those region servers at any location without regard to WAN latency.
If you’ve ever struggled with HBase performance, you should give Non-Stop for Apache HBase a close look.