Hyperconverged infrastructure gets a Hyperbenchmark
VMware, Nutanix and even Oracle played nice to cook up TPCx-HCI
The Transaction Processing Performance Council (TPC) has released a benchmark for hyperconverged infrastructure.
The new TPCx-HCI benchmark is a derivative of the TCP-V benchmark for database performance on virtual machines. The Council's cooked up the benchmark after deciding hyperconverged infrastructure has sufficient momentum that buyers want to be able to compare apples to apples. The benchmark therefore assumes that the tin it will run on integrates compute, storage and networking – the generally-agreed definition of hyperconverged.
TPCx-HCI exercises that tin in two ways. One runs an “elastic workload that varies the load delivered to each of the VMs by as much as 16x, while maintaining a constant load at the cluster level.” The workload is designed so that it will do best when VMs are migrated across nodes in the cluster, in order to test VM migration and data access speeds.
There's also a “data accessibility test” in which “a node is powered down ungracefully, and the benchmark continues to run on the other nodes. The test sponsor has to include a throughput graph for this test, demonstrating the impact on performance, as well as report the recovery time to regain resilience.”
TCPx-HCI was created in just seven months, an unusually short amount of time, with Cisco, DataCore, Dell, HP Enterprise, Huawei, Microsoft, Nutanix, Oracle, Red Hat and VMware all helping.
Oracle's an odd one on that list, as it offers converged but not hyperconverged systems. The Register assumes that Big Red chose to play because knows its databases and apps will run on hyperconverged kit, rather than planning a product itself given its focus on cloud. But you never know!
Notable absentees are IBM and NetApp. The former's absence isn't a huge issue, given its VersaStack is done jointly with Cisco. The latter was late-ish to the hyperconverged party, so perhaps just missed the boat.
Now to grab some popcorn and see if the various hyperconverged players can find fault with their rivals results, as often happens once benchmarks reach the real world. ?