MongoDB 3.x Performance Secrets: Benchmarks, Issues, and Best Practices

This article shares MongoDB performance insights, test results, and practical recommendations based on real‑world experience.

Standalone Test

Benchmark on a 24‑core machine with PCIe SSD using MongoDB 3.0.4, snappy compression, journaling enabled, and write concern w=1.

<code>3.0.4 / snappy / standalone / w=1 / enable journal / 1k per document</code>

Results:

<code>[OVERALL], Throughput(ops/sec), 118253.63038645286
[UPSERT], AverageLatency(us), 496.5021718
[UPSERT], 99thPercenTleLatency(ms), 0</code>

Replica Set Test

Same hardware, replica set configuration.

<code>3.0.4 / snappy / replicaTon / w=1 / enable journal</code>

Results:

<code>[OVERALL], Throughput(ops/sec), 68947.44825494009
[UPSERT], AverageLatency(us), 844.9684893
[UPSERT], 99thPercenTleLatency(ms), 0</code>

Read‑Write 95:5 Scenario

<code>3.0.4 / snappy / replicaTon / w=1 / enable journal read:write=95:5</code>

Results:

<code>[READ], AverageLatency(us), 394.409190121636
[READ], 99thPercenTleLatency(ms), 0
[UPDATE], AverageLatency(us), 562.6090867394253
[UPDATE], 99thPercenTleLatency(ms), 0</code>

Checkpoint/eviction algorithms are still being improved; MongoDB 3.0.6 shows greater stability.

Disk Space Savings

With WiredTiger compression, 1.02 billion documents occupy only 270 GB, about 8 % of the space required by the 2.6 version (3.12 TB for 1.18 billion documents), dramatically reducing SSD/PCIe costs.

Version‑Specific Issues

3.0.2 Issues

Key JIRA issues:

WT‑1941 / SERVER‑18629 – unnecessary fsync of data directory; fixed in 3.0.3. Workaround:

<code>log=(xxx,path=journal/,xxx)    // add trailing slash after journal</code>

SERVER‑18314 – recommendation to use XFS filesystem.

3.0.4 Issues

Fixed previous issues but introduced memory‑leak problems related to cursor handling.

3.0.6 Issues

Fewer severe problems; many issues were quickly resolved (e.g., WT‑2104).

Driver Issues

An async driver failed to detect primary after failover in replica set tests; thorough testing of new drivers and failover scenarios is essential.

Application Issues

Inconsistent time formats and mixed field types caused query errors. Recommended solutions:

Enforce application‑level schema constraints.

Use tools like variety to detect field type diversity.

Apply table‑level privileges.

Maintain consistent metadata and avoid large documents, overly long field names, and unnecessary indexed arrays.

Connection Issues

Observed half‑open connections from mongos; adjusting keep‑alive settings helped mitigate the problem.

Our Recommendations

Standardize data schema metadata.

Avoid indexed arrays when updates are frequent.

Observe naming restrictions.

Prevent large documents.

Keep field names concise.

Pay attention to TTL timing.

Read MongoDB documentation extensively and stay engaged with the community.

Best Practices – Driver

Set query timeouts (similar to MySQL

statement_timeout

) to protect the server from hanging long queries.

Since NoSQL lacks SQL tuning, deep dive into schema and architecture design, and perform code reviews.

Best Practices – Application

Avoid negative predicates (e.g.,

not in

,

!=

).

Update only necessary fields instead of reading and rewriting whole documents.

Use

explain

to understand query plans.

Choose appropriate read preferences based on workload.

When sharding, select a shard key that balances write load while supporting queries.

Limit each MapReduce job to a single core.

Heterogeneous Replication

Combining technologies to replicate MySQL data into MongoDB for high‑throughput use cases, using a custom VDP solution that parses binlogs and writes to MySQL, Kafka, HBase, MongoDB, etc., via a drag‑and‑drop UI.

Another Scenario

For reporting, extract MongoDB data to HDFS and query with Hive/Presto. Include a

lastupdatetime

field to enable incremental extraction and reduce load on MongoDB.

Automation and Monitoring

MongoDB 3.x provides over 300 monitoring metrics; we collect and visualize them with Zabbix, enabling quick issue detection. Deployment automation is handled with Ansible.