In today’s topic, I’d like to talk about the Meltdown and Spectre vulnerabilities. But not about the harm they cause, this has been covered widely in numerous articles, but how Microsoft patches intended to protect you from the vulnerabilities, affect (if they do) the hardware performance. Before we take a deep dive into the tests and numbers, let me tell a few words about Meltdown and Spectre and outline the testing scope to make sure we speak one language.
Performance or protection? How Microsoft patches against Meltdown and Spectre influence CPU, RAM and Disks performance
For quite a long time, System Center Virtual Machine Manager (SCVMM) has a feature called Dynamic Optimization. Its main goal is to automatically rebalance VMs between the participating cluster nodes in case the placement is unequal. Now, this feature has partially became available in Windows Server 2016 in the form of Node Fairness. It balances the workloads among the hosts in a Hyper-V Failover Cluster and automatically live migrates guests from an overloaded node to a less busy one with zero downtime.
Node Fairness goes embedded in Windows Server 2016 and is intended for deployments without SCVMM. SCVMM Dynamic Optimization delivers more versatile functionality than Node Fairness. Regarding this fact, Dynamic Optimization is recommended for balancing workloads among the cluster hosts. However, to use this feature, you need an additional license from the main operating system.
Now that we know what Node Fairness is, let’s take a look at how this service works.
In the previous article, I’ve measured the performance of NFS vs iSCSI to find out which network protocol is faster as a storage for virtual machines on VMware ESXi. Well, iSCSI beats NFS under all testing patterns. Additionally, I’ve evaluated and compared the performance of NFS client connected to Linux (Ubuntu Server 17.10 distributive) and to Windows Server 2016. According to the results, NFS server performance on Linux was higher than that on Windows.
In the previous parts, I’ve shown you the process of configuring NFS and iSCSI protocols between our servers. So now, we’ve got everything ready for running our performance tests and finally finding out which network protocol is faster as a storage for virtual machines on VMware ESXi: NFS or iSCSI.
So to benchmark the iSCSI performance, I’ve created the StarWind device on the server and connected it to the ESXi host over the iSCSI protocol. As to OS for running further tests, I’ve used Windows Server 2016.
Subscribe to our posts
- How is NVMe-oF doing? Part 1: Linux NVMe-oF Initiator + Linux SPDK NVMe-oF Target
- Setting up a Windows Failover Cluster for a home lab
- Performance comparison: SQL Server Failover Cluster Instance on S2D vs. SQL Server Basic Availability Groups on Storage Spaces. The battle starts
- Can SQL Server Failover Cluster Instances run on S2D twice as fast as SQL Server Basic Availability Groups on Storage Spaces? Part 2: Studying FCI performance
- Can SQL Server Failover Cluster Instance run on S2D twice as fast as SQL Server Basic Availability Groups on Storage Spaces? Part 1: Studying BAG performance