Apache HBase

Following are the key areas of differences between YugabyteDB and Apache HBase.

Simpler software stack

Apache HBase relies on HDFS (another complex piece of infrastructure for data replication) and on Apache Zookeeper for leader election, failure detection, and so on. Running HBase smoothly requires a lot of on-going operational overheads.

High availability and fast failover

In HBase, on a region-server death, the unavailability window of shards/regions on the server can be in the order of 60 seconds or more. This is because the HBase master first needs to wait for the server's ephemeral node in Zookeeper to expire, followed by time take to split the transaction logs into per-shard recovery logs, and the time taken to replay the edits from the transaction log by a new server before the shard is available to take IO. In contrast, in Yugabyte, the tablet-peers are hot standbys, and within a matter of few heartbeats (a few seconds) detect failure of the leader, and initiate leader election.

C++ implementation

  • Avoids GC tuning; can run better on large memory machines.
  • Richer data model: YugabyteDB offers a multi-model/multi-API using YSQL, YCQL.
  • Rather than deal with just byte keys and values, YugabyteDB offers a rich set of scalar (int, text, decimal, binary, timestamp, etc.) and composite types (such as collections, UDTs, etc.).

Multi-data center deployment

Flexible deployment choices across multiple data centers or availability zones. HBase provides strong consistency only within a single data center and offers only an asynchronous replication alternative for cross-DC deployments.