9:14in productionCh. 1 · Not SQL. Not relational./ 9:14 · ceiling 15 min
Software · Systems
Apache HBase
2006
HBase delivers Bigtable’s speed on Hadoop’s bones—but only if your data is sparse, your consistency needs are strict, and your tolerance for operational complexity is high.
HBase is a wide-column, non-relational, distributed database modelled after Google's Bigtable. It runs on HDFS or Alluxio. It provides fault-tolerant storage for large sparse datasets with fast read/write operations, high throughput, low I/O latency, and CP consistency per the CAP Theorem. It features compression, in-memory operation, and Bloom filters on a per-column basis. Tables can serve as input and output for MapReduce jobs and are accessible via Java API, REST, Avro, or Thrift.
HBase is a wide-column store, not a relational database—and its lineage to Bigtable defines its structure and limits.
2:32
Built on Hadoop’s foundation
It runs on HDFS or Alluxio, and that dependency shapes its performance: fast random access, but only atop a batch-optimised file system.
4:12
Optimised for sparsity
Compression, Bloom filters, and in-memory operation make sparse-data lookups efficient—but they do not fix write amplification or compaction overhead.
5:59
CP, not AP
It supports MapReduce and multiple APIs—but sacrifices availability to guarantee consistency, making it unsuitable for highly partitioned environments.
Worth your time?
Yes. Study the whole thing.
3.5/ 5
What works
fast random reads/writes on sparse data
MapReduce integration
per-column optimisations
CP consistency
What does not
multi-row transactions
SQL
ACID
schema enforcement
Study it if
engineers managing sparse, high-volume time-series or log data in Hadoop environments
Skip it if
teams needing ad-hoc SQL analytics, strong consistency across regions, or simple operational models
The written brief1 min read
What it is and the problem it solves
HBase is a non-relational, wide-column distributed database modelled after Google’s Bigtable. It solves the problem of fast random read/write access to sparse, massive datasets—like finding rare non-zero values in billions of mostly empty records.
How it works
HBase stores data in wide-column tables across distributed nodes. It runs on HDFS or Alluxio. It uses compression, in-memory operation, and per-column Bloom filters. It provides CP consistency under the CAP Theorem.
What works
Fast read/write operations on large datasets. High throughput and low I/O latency. Fault-tolerant storage of sparse data. Integration with MapReduce. Access via Java API, REST, Avro, and Thrift.
What does not
HBase does not support multi-row transactions or relational joins. It does not guarantee availability during network partitions. It does not offer native SQL or schema enforcement.
What it changes
HBase changes how large-scale sparse data is accessed in batch-adjacent systems. It enables real-time querying atop Hadoop’s batch infrastructure—but only for specific access patterns.
Is it worth your time
Yes—if you need fault-tolerant, low-latency random reads/writes on sparse, massive datasets within a Hadoop ecosystem. No—if you require ACID transactions, SQL, or eventual consistency.
