×

Expanding Enterprise Computing Options with an All-Arm Cluster – The Hammerhead Way

Suse Linux

Expanding Enterprise Computing Options with an All-Arm Cluster – The Hammerhead Way


Background:

The Hammerhead Consortium leveraged its Ampere Computing-powered cluster in its latest test iteration showcasing a ScyllaDB environment in an all-Arm cluster. The results will be made available in a joint session between Ampere Computing and ScaleFlux at the upcoming ScyllaDB Summit 2023.

The objective was to show the ability to deliver high performance numbers with lower power consumption due to the higher CPU density available with Ampere’s Altra and ScaleFlux’s embedded intelligent computational storage technology.

The Hammerhead Consortium consists of technology and business professionals from Ampere Computing, Arm, Micron, NVIDIA, and SUSE working together to expand the ARM64 application eco-system by testing and validating select compute workloads while holistically improving their overall performance leveraging the product components brought by the consortium.

In other words: “We show a path to what’s possible”.

Test Configuration:

A subset of the Hammerhead III cluster was used for the testing.  3 nodes were configured to act as Scylla Cluster Servers with ScaleFlux Storage Accelerators added to them.  Another 3 nodes (without the Storage Accelerators) were used to act as client test units.

(Illustration provided by Ampere Computing)

Tests, Results and Findings:

To test the performance of our cluster we used cassandra-stress benchmark utility. Three different workload profiles were tested – 100{491a634df86f69b1a2c85c603889f03dfb83f91ec539f453861be7aa79511ef2} Read and 75{491a634df86f69b1a2c85c603889f03dfb83f91ec539f453861be7aa79511ef2}/25{491a634df86f69b1a2c85c603889f03dfb83f91ec539f453861be7aa79511ef2} Read/Write across 10 Billion records to demonstrate performance from disk and 50{491a634df86f69b1a2c85c603889f03dfb83f91ec539f453861be7aa79511ef2}/50{491a634df86f69b1a2c85c603889f03dfb83f91ec539f453861be7aa79511ef2} Read/Write across 1 million records to show performance when accessing data in memory.

For each of these tests the three node cluster performed very well, serving over 1 million requests/sec and maintaining P99 write latencies under 8 milleseconds. The 75/25 R/W test serviced 1.1 million requests/sec and the other two tests served 1.4 million requests/sec.

Key Takeaways/Call To Action:

  • The Hammerhead consortium continues to demonstrate the benefits of the combined product ecosystem.

    • A complete arm64-based, hardware and software stack ready to support applications at-scale.

    • High-performance, density, and efficiencies are reachable with the combined product ecosystem.

  • ISVs and Developers alike who want similar benefits can take advantage of the Ampere Developer Access Program. Gain remote access to bare metal servers, trial systems shipped to you or access to partner cloud resources – depending on your requirements.

  • ISVs looking to gain access to all the elements of the SUSE stack and certify their applications can join the SUSE One Partner program.

(Visited 1 times, 1 visits today)



Source link