Summary

The introduction of new server technologies allows customers to use the new functionality to deploy solutions. It can also provide an opportunity for them to review their current infrastructure to see whether the new technology can increase efficiency. With this in mind, Dell Technologies recently conducted performance testing of an Elasticsearch solution on the new Dell PowerEdge R760 and compared the results to the same solution running on the previous generation R750 to determine whether customers could benefit from a transition. All testing was conducted in Dell Labs by Intel and Dell engineers in April 2023.

Choosing which CPU to deploy with an advanced solution like Elasticsearch can be challenging. A customer looking for maximum performance would typically start with the most expensive CPU available, while another customer might make a choice that offers a tradeoff between performance and price. For the purposes of this test, we decided to benchmark the new R760 with a lower cost processor so that we could compare the results to a previous generation R750 server using the top end Intel® Xeon® Platinum 8380 CPU.

Workload overview

An Elasticsearch solution includes multiple key components that combine into the “Elastic Stack”.

• Elasticsearch: RESTful, JSON-based search engine
• Logstash: Log ingestion pipeline
• Kibana: Flexible visualization tool
• Beats: Lightweight, single purpose data shippers

Methodology

To conduct the testing, we deployed Rally 2.7.1 as the benchmarking tool. Using an OpenShift Kubernetes cluster, each server was configured to create an Elasticsearch cluster with eight instances (containers). Next, each system ran 10 cycles of searches to establish a “steady-state” flow of data as an indexing test. The performance of each system was measured by capturing the mean throughput of the bulk index (doc/s) and the search query latency (ms).

The benchmark simulated storing log files (application, http_logs, and system logs) and users who use Kibana to run analytics on this data. The test executes indexing and querying concurrently. Data replication was enabled, and software configuration was the same on both platforms.

The average CPU utilization during the test was 80%.

Dataset

Logging - server log data

The logging-indexing-querying workload generates multiple server logs before the test. The benchmark executes indexing and querying concurrently. Queries were issued until indexing was complete.

We used the following log types:

• Nginx access and error logs
• Apache access and error logs
• Mysql slowlog and error logs
• Kafka logs
• Redis app logs
• System syslog logs
• System auth logs

Who uses it? This data is typically produced by web services and could be used to validate HTTP responses, track web traffic, and monitor databases and system logs.

Hardware configurations tested

Note: The Dell Ent NVMe P5600 MU U.2 3.2TB Drives are manufactured by Solidigm.

Recommended customer pricing for the CPUs used in the tested configurations

• R750 - Intel Xeon Platinum 8380 - $9,359 - reviewed on June 6, 2023
• R760 - Intel Xeon Platinum 8460Y+ - $5,558 – reviewed on June 6, 2023

Price Delta:

Sources:

8380: Intel Xeon Platinum 8380 Processor 60M Cache 2.30 GHz Product Specifications

8460Y: Intel Xeon Platinum 8460Y Processor 105M Cache 2.00 GHz Product Specifications

Software configuration

Test results

The following results represent the mean of 10 separate test runs.

Indexing Throughput (docs/s)

Indexing throughput indicates how many documents (log lines) that Elasticsearch can index per second.

Note: Higher is better

Latency Improvement

Latency improvement indicates how much faster search query results return.

Note: Higher is better

Power consumption and calculations

Conclusion

Choosing the right combination of server and processor can increase performance, reduce latency, and reduce cost. As this testing demonstrated, the Dell PowerEdge R760 with 4th Generation Intel Xeon Platinum 8460Y CPUs was up to 1.24x faster than the Dell PowerEdge R750 with 3rd Generation Intel Xeon Platinum 8380 CPUs.

An important element to consider is that the R760 was able to accomplish all of this using CPUs with a recommended customer price that was more than 40% less, thus reducing capital expense. The testing further demonstrated that customers can reduce operating costs by implementing new technologies that can deliver more work per watt.