This week SPEC has published a new SPEC CloudTM IaaS 2016 result for a private cloud configuration built using VMware vSphere 6.5 and VMware Integrated OpenStack 3.1 (VIO 3.1) and Dell PowerEdge Servers. Working with VMware, Dell has pushed their lead in cloud performance even further. This time, the primary metric produced was a Scalability score of 78.5 @ 72 Application Instances (468 VMs). The Elasticity score was 87.4%.
VMware and Dell are active participants in SPEC and have contributed to the development of its industry standard benchmarks including SPEC Cloud IaaS 2016. Both organizations strongly support SPEC’s mission to provide a set of fair and realistic metrics on which to differentiate modern systems and technologies.
The submission for the new high-water mark was a result of a cooperative effort between the VMware Performance Engineering team and the Dell Solutions Performance Analysis team. The result has undergone two weeks of peer review by the members of the SPEC Cloud subcommittee prior to its acceptance and publication on the SPEC.org website.
SPEC Cloud IaaS 2016 can be used to benchmark both public and private clouds. SPEC Cloud takes two popular Big Data applications, Cassandra and Hadoop, and puts them at the core of two distributed multi-instance workloads, the Yahoo Cloud Serving Benchmark (YCSB) and the HiBench K-means benchmark, respectively. The YCSB workload represents I/O intensive social media applications using NoSQL transactions. The K-means workload represents CPU-intensive machine learning and data mining applications. The benchmark requires a cloud framework and API to manage the cloud functions and provide the interface for the benchmark harness (CBtool). Instances may be created using virtualization, bare-metal, or containers.
The primary metrics are Scalability – a normalized score that sums the work completed by all AIs; and Elasticity – a percentage, measuring how, under continually increasing load, each AI performs during the test compared to the baseline taken when only a single AI is run.
During the main test phase of the benchmark, these multi-instance workloads – called Application Instances (AIs) are injected into the Cloud under test every few minutes. While the new AI’s are provisioned, all previously injected AIs continue running their application and meeting QoS until the end of the test run.
As each new AI is provisioned and configured, the AI starts a cycle of generating a new data set and then using that data in the context of its workload. CBtool monitors the status of all AIs and workloads ensuring that they meet their QoS criteria as it continues to request new AI’s be deployed. The YCSB throughput and K-means completion times for each cycle in each AI is collected along with the provisioning times to calculate the overall metrics at the end of the test.
In this test, we used vSphere’s ESXi 6.5 as the hypervisor on all servers and utilized VIO 3.1 as the cloud framework. The OpenStack API adapter included with the benchmark was used by CBtool to drive load to the private cloud set up for the test. The cloud’s compute nodes consisted of two DRS Clusters, each with four ESXi hosts. A third cluster was used to host the VIO 3.1 deployment. Each cluster was built using a Dell PowerEdge FX2 Chassis with 4 FC630s. A fourth cluster of Dell PowerEdge C6220s, was used to host the vCenter Server Appliance, the CBtool Harness VM, along with a third VM used to access the OpenStack Horizon GUI. The diagram for the hardware configuration from the FDR is shown below:
VMware technology is the foundation for best-in-class private clouds deployed by more than 500,000 customers worldwide. This benchmark is yet another demonstration that VMware technology can support virtually any workload — from traditional enterprise apps to modern container-based microservices – or in this case, big data applications running on private cloud built on vSphere 6.5 and VIO 3.1 and Dell PowerEdge servers. VMware Integrated OpenStack (VIO) and vSphere Enterprise Plus can accelerate deployment and management of a production-grade OpenStack and provide vendor-neutral APIs to get more out of your Software-Defined Data Center (SDDC).
For more details on the SPEC CloudTM IaaS 2016 benchmark visit www.spec.org/cloud. To see the full disclosure report for this test visit www.spec.org/cloud_iaas2016/results/res2017q2/.