Hyper-Converged Reviews
by StorageReview Enterprise Lab

Dell EMC VxRack Node Powered By ScaleIO Review: Wrap-Up

Over many months, StorageReview has been testing and analyzing two Dell EMC VCE VxRack Node all flash PF100 clusters – which is the predecessor to the recently released Dell EMC ScaleIO Ready Node. One set of tests centered on the system configured in HCI, and the other in two-layer. If this sounds a bit strange, it is, as no storage is more flexible than ScaleIO. Hyperconverged is typically that; compute and storage combined into the same easy-to-manage platform. And just as it sounds two-layer is the separation of storage and compute as is typical in a traditional architecture. Dell EMC ScaleIO eschews the strict definition of both notions, allowing an enterprise to configure either way, or both ways. The Dell EMC ScaleIO solution meets the demands of organizations that want the best of each world, whenever they like.

Read more
by StorageReview Enterprise Lab

VMware VSAN 6.2 All-Flash Review

Last year we published a detailed series of reviews on VSAN 6.0 in a hybrid configuration. In February of this year, VMware launched VSAN 6.2, which brought data reduction via deduplication and compression, as well as a number of other features to the market. Data reduction is a major step forward for VSAN as it piggybacks on the massive surge in flash adoption, thanks in large part to falling prices and high-quality options. When VSAN is enabled in an all-flash configuration, users taking advantage of data reduction can effectively triple (or more) their capacity depending on workloads, which really brings the value of flash to everyone from ROBO to the enterprise. 

Read more
by StorageReview Enterprise Lab

EMC VxRack Node Powered By ScaleIO: Scaled Sysbench OLTP Performance Review (HCI)

Scaled Sysbench is the final performance test of the all-flash EMC ScaleIO VxRack Node configured in HCI. In two-layer, we pressed the system to 99.2% capacity and saw phenomenal throughput with the four Dell servers acting as load gens. This time, with the compute and storage together in the 2U chassis, we run the same testing to see what happens. We'd expect to see more weight on the CPUs as the limiting factor, as the system has more work to, but thus far in the HCI testing, ScaleIO has proven to be extremely efficient and lightweight from an overhead perspective; something that is generally the reverse in other HCI solutions.

Read more
by Kevin OBrien

EMC VxRack Node Powered By ScaleIO: VMmark Performance Review (HCI)

When deployed in hyperconverged mode, the EMC VxRack Nodes running ScaleIO are often found running VMware. While not always the case, the nodes support just about everything, VMware is the easiest to understand from a performance perspective thanks to VMmark. When we reviewed ScaleIO running VMmark in two-layer, the all-flash VxRack Node handled 26 tiles. The system could have probably handled a few more as it had capacity to spare, but the four loadgen servers were at their max. This time we run the same benchmark in HCI, combining the storage and compute into one 2U box versus the 10U total in the two-layer test. In this test the efficiency of the management software will be obvious as the box will be at saturation points everywhere; storage capacity, CPU utilization and RAM footprint will all be under duress. 

Read more
by StorageReview Enterprise Lab

EMC VxRack Node powered by ScaleIO: SQL Server Performance Review (HCI)

After our first look at EMC's VxRack Node running in HCI mode measuring synthetic workload performance, we turn our attention to SQL Server OLTP performance. In this particular test our focus is on application-level latency, instead of completely stressing the system in terms of both CPU and storage I/O. To find out just how well ScaleIO holds up under pressure, we put its HCI configuration head to head against its phenominal two-layer results to see if EMC still has a few tricks up its sleeve.

Read more
by Brian Beeler

EMC VxRack Node powered by ScaleIO Review: Synthetic Performance Review (HCI)

We've spent a good deal of time talking about EMC ScaleIO's capabilities in a two-layer, or storage online configuration. In this next series of the review, we break out results with ScaleIO set up as a hyper-converged system. It should be noted that ScaleIO can actually work in a third mode where it combines both of these options. Fundamental to the HC setup, though, is the fact that both the SDS (ScaleIO Data Server) and the SDC (ScaleIO Data Client) run in the same environment. In this configuration the applications and storage share the same compute resources. In our case, that means the VxRack Node's high-performance 2U 4-node PF100 chassis that resides in our lab. 

Read more
by StorageReview Enterprise Lab

VMware Virtual SAN Review: HCIbench Synthetic Performance

One of the most common ways to stress a storage platform is through synthetic workloads or four-corners type of testing. In this testing scenario, we look at the peak I/O a platform can generate, as well as the peak bandwidth. While these figures aren't supposed to be used in place of application tests, they are helpful in comparing one platform to another and provide valuable insight for customers working on a PoC. To that end we have migrated hyper-converged synthetic tests to HCIbench, which VMware released earlier this year. At its core HCIbench leverages the industry-accepted VDbench workload generator and distributes VMs across a target cluster and aggregates the results. We use our own I/O profiles for this tool, which are linked below for users to replicate our tests.

Read more
by StorageReview Enterprise Lab

VMware Virtual SAN Review: Scaled Sysbench OLTP Performance

After looking at the performance levels of the VMware VSAN cluster with a traditional Sysbench OLTP workload, we wanted to find out how well the platform responds with an increased workload for more demanding use cases. The initial deployment was four Sysbench VMs, 1 per node, but that workload didn't bring the disk I/O to a high enough range where we felt resources were being fully utilized. This is similar to a customer running a POC, testing it under a subset of their current workload, but not measuring how well the platform responds as workloads grow over time or as more application data is migrated over. To better understand how this VSAN cluster responds under ever-increasing MySQL workloads, we scaled the four Sysbench VM (1 per node) benchmark up to 8 and 12 total VMs.

Read more
by StorageReview Enterprise Lab

VMware Virtual SAN Review: SQL Server Performance

Continuing on the path of VMware VSAN performance testing, the next test looks at Microsoft SQL Server TPC-C running across the cluster. This test uses SQL Server 2014 running on Windows Server 2012 R2 guest VMs, being stressed by Dell's Benchmark Factory for Databases. While our traditional usage of this benchmark has been to test large 3,000-scale databases on local or shared storage, in this iteration we focus on spreading out four 1,500-scale databases evenly across our VSAN cluster to better illustrate the aggregate performance of the 4-node cluster. 

Read more
by StorageReview Enterprise Lab

VMware Virtual SAN Review: Sysbench OLTP Performance

To measure the performance of the VMware VSAN cluster in transactional database workloads, we first leverage the Sysbench OLTP benchmark, paying close attention to total aggregate performance. The Sysbench OLTP benchmark runs on top of Percona MySQL leveraging the InnoDB storage engine operating inside a CentOS installation. While a traditional SAN infrastructure can better cope with large single workloads, hyper-converged systems are designed to spread that load across all nodes in the system. To that end, we deployed four Sysbench VMs on the VSAN cluster, 1 per node, and measured the total performance seen on the cluster with all operating simultaneously.

Read more