Today, we are happy to a new YCSB performance results with Couchabse Server 4.5 running on Tegile storage arrays. This benchmark shows the combined benefits of linear scalability of Couchbase Server as well as powerful storage subsystem provided by Tegile. 

Introducing Tegile IntelliFlash Storage Array

Tegile IntelliFlash storage arrays deliver low latency and high IOPS for virtual server environments running performance-hungry applications like Couchbase. By seamless integrates multiple grades of storage media, Tegile Intelligent Flash Storage Array is able to reduce Couchbase write time dramatically. Besides that, Tegile storage array is able to reduce storage footprint and maximize storage efficiency through inline compression and deduplication.

Introducing Couchbase Server 4.5

We introduced Couchbase Server 4.5 in June 2016. Couchbase Server 4.5 is a significant release that delivers superior performance. We enhanced KV functionality and performance, introduced memory optimized index and improved query optimization. A 3rd party company published a YCSB benchmark between Couchbase and MongoDB and the result showed that Couchbase outperformed MongoDB by 7X on reads, 5X on writes and 3X on queries. You can find the full report here.

Couchbase/Tegile YCSB Benchmark Result

As a database company, we are always interested in seeing how our software runs on various storage arrays. With the innovation of Tegile storage array, we are excited to see how Couchbase Server performs on this platform.

We performed this benchmark on the following server architecture:

As illustrated in the figure above, this architecture contains:

Hardware:

  • Server
    • 1x Cisco UCS 5108 chassis
    • 8x Cisco UCS B series blade, as the cluster to deploy Couchbase Server
  • Storage
    • 1x Tegile IntelliFlash storage array
    • Storage provisioning on the IntelliFlash storage array:
      • 1x iSCSI LUN for SAN boot of the UCS blade server
  • Network and Interconntect
    • 2x Cisco UCS FI (Fabric Interconnects)
    • The UCS chassis and the IntelliFlash arrays are connected directly through UCS FI in No-Single-Point-of-Failure fashion
    • Interconnect protocols: 10GbE iSCSI and SMB 3.0
  • Total 9 RU rack space

Software:

  • 8x Operating System on each blade: Linux – CentOS 7.1
  • 8x Couchbase Server 4.5: Couchbase is deployed on all of the Cisco UCS blades.

YCSB Benchmark Results:

The YCSB benchmark was run with 4 Cisco UCS blades, Tegile IntelliFlash and Couchbase Server 4.5 to verify the reference architecture.

YCSB Workload A

This is a mixed workload with 50/50 reads and writes. Session stores, sensor management are all example applications for this workload. With 4 YCSB clients, Couchbase Server maintained 95th percentile latency of 0.24ms (240 microseconds) for each operation. The results show linear scalability with Couchbase Server 4.5 running on Cisco UCS and Tegile IntelliFlash.

Figure 2 – YCSB Workload A  (X-Axis: Number of YCSB clients)

YCSB Workload E

All of the workload queries short range of documents/rows, instead of individual documents/rows.Threaded conversations, listing customers by postal code, product by category are all examples for this workload.With 4 YCSB clients, Couchbase Server 4.5 maintained 95th percentile latency of 13.64ms for each operation. Each operation is doing an index scan by range, ordering and getting the top-k documents.  That means, each workload operation is doing many more operations, just like a complex query.

                                                                           Figure 3 – YCSB Workload E  (X-Axis: Number of YCSB clients)

Conclusion

The YCSB results above show consistently low latency at high throughput for the two typical workloads. The combination of Couchbase database, Cisco UCS Common Platform Architecture combined with the flexibility of Tegile IntelliFlash makes it a dependable enterprise solution for high performance or high capacity workloads.

Recognition

This work is done by colloborated efforst among Tegile, Cisco and Couchbase. Special thanks go to Keshav Murthy, Ramesh Chitor, Raghu Nambiar and Harrison Waller. You can find the full solution brief here.

Author

Posted by Qi Zhu, Product Manager, Couchbase

Qi Zhu is a Technical Product Manager at Couchbase. Check out his 4 publications in distributed computing and software reliability.

Leave a reply