Can TokuDB Really Cut Storage by 14× and Boost Inserts? Real‑World Test Results

Facing ever‑growing data volumes, the author evaluated the open‑source TokuDB storage engine (released in 2013) to address capacity constraints, testing three advertised features: high compression (claimed 1:12), high insert performance (claimed at least 9× InnoDB), and fast online addition of indexes and columns.

Compression Test

The original database size of 1126 GB was reduced to 76 GB after migrating to TokuDB, achieving a 14× compression ratio.

Insert Performance

A replication‑sync test showed that, without changing configuration or adding multithreaded replication, TokuDB’s sync speed was about 1.9× faster than InnoDB.

Online DDL Operations

Using a 425 MB table on a RAID‑5 SSD server, the following observations were made:

Adding an index: InnoDB faster, TokuDB slower.

Dropping an index: InnoDB fast, TokuDB instantaneous.

Adding a column: InnoDB slow, TokuDB instantaneous.

Dropping a column: InnoDB slow, TokuDB instantaneous.

TokuDB therefore shows a clear advantage for online DDL, especially for column operations.

Sample Table Definition

CREATE TABLE `timeline_1009` (
  `uid` bigint(16) unsigned NOT NULL,
  `vflag` tinyint(4) unsigned NOT NULL DEFAULT '0',
  `status_id` bigint(16) unsigned NOT NULL,
  `source` int(6) unsigned NOT NULL DEFAULT '0',
  `fflag` int(6) unsigned NOT NULL DEFAULT '0',
  `mflag` tinyint(4) NOT NULL DEFAULT '0',
  PRIMARY KEY (`uid`)
) ENGINE=TokuDB DEFAULT CHARSET=utf8 ROW_FORMAT=TOKUDB_LZMA;

Operation Timing Comparison

InnoDB operation logs (seconds): create index 36.48, drop index 0.29, add column 28.64, drop column 27.29.

TokuDB operation logs (seconds): create index 56.56, drop index 0.05, add column 0.01, drop column 0.00.

InnoDB’s add‑column operation creates a temporary table, causing massive row‑affected counts, whereas TokuDB performs the change in the background using a Fractal‑tree structure, avoiding such overhead.

Resource Consumption

CPU Usage: TokuDB consumes roughly twice the user‑mode CPU compared to InnoDB.

IOPS: Write IOPS for InnoDB are about 5.5× higher than TokuDB; read IOPS are comparable.

Response Time

Despite the smaller database size (TokuDB 61 GB vs InnoDB 743 GB), TokuDB’s average response time is about 2.2× higher, likely due to decompression overhead.

Conclusion

Advantages of TokuDB: high compression ratio, superior insert performance, near‑instantaneous online DDL.

Disadvantages of TokuDB: higher CPU usage in user mode and longer response times.

Overall, TokuDB’s features are attractive for solving storage‑related challenges, but its trade‑offs require careful evaluation before production deployment.