In virtual environments, VMware vSphere could restrain the maximum performance of the system infrastructure if it is not properly optimized. VMware provides great documentation that helps understand the design considerations and general deployment options for deploying vSphere infrastructures, however some of these considerations differ for modern All-Flash Array.
All-Flash Array are fast, very fast; yet it is common to see performance limitations when Proof of Concepts are configured and tested under these storage systems and VMware environments. To avoid performance limitations and needless fine-tuning, this guide highlights the key considerations and configuration settings that promote a high-performance and reliable virtualized environment for connecting vSphere assets and clusters with the AccelStor NeoSapphire™ platform.
These 7 tips have been taken from the AccelStor NeoSapphire™ AFA Best Practices for VMware document, where you can find the full set of recommendations and configurations.
1. Use multiples SAN switches for redundancy One common mistake of testing storage systems in a VMware environment is using a direct connection to ESXi hosts, or just using a single switch. Although many AFAs supports to be directly attached to the hosts, always use a SAN switch to perform POCs. At least two, for production environments. Multiples switches increase the number of data paths between the storage array and the ESXi hosts, and with this, performance and data redundancy.
2. Use all available Data Ports As simple as it may sound, always use all the available data ports from the storage array to interconnect the ESXi hosts. Storage data ports are usually available in both, Fiber Channel (16Gbps) or iSCSI (10Gbps). Depending on the used protocol, a single port will tolerate an average of 250K IOPS @4KiB. When connecting All-Flash Array to the infrastructure, using a single data port (or not the total of them) is a mistake, this constrains the maximum performance that the systems are capable of offering.
3. Avoid the use of Jumbo Frames For iSCSI protocols, Jumbo Frames might improve the network performance by making data transmissions more efficient; the condition, each link in the network path -- storage, SAN switch and ESXi hosts – must be configured to enable jumbo frames. Missing one of these links will lead to inconsistent performance and packet loss. In All-Flash Array, Jumbo Frames offers marginal performance gains while adding extra complexity to the infrastructure. Tip: avoid Jumbo Frames.
4. Change the Path Selection Policy to Round Robin To implement load balancing across the physical paths available to your ESXi host, Active-Active All-Flash Array requires to change the VMware path selection policy to Round Robin. The goal is to optimize performance regarding throughputs, such as I/O per second, megabytes per second, or response times by using all available paths.
5. Do not modify queue depths Performance-wise, queue depths play an important role through VMware environments, and usually, the decision to modify them is to achieve comparable performance with non-virtualized environments. The key thing understand about queue depths is that the ESXi host is designed to provide some level of fairness, equally sharing resources among the virtual machines, datastores, etc. Similar to Jumbo Frames, queue depths add complexity to the infrastructure, so instead of altering queue depths, is better to increase the number of Virtual Machines and datastores. Also, it is important to keep in mind that more of one ESXi host is needed to match the full performance of an All-Flash Array.
6. Ensure that VAAI is Enabled VAAI offloads IO-intensive processing from the ESXi environment to the All-Flash Array, enabling the delivering of more efficient, agile, and cost-effective VMware environments. The key of VAAI is to accelerate the provisioning, cloning, and migrations of Virtual Machines while keeping low computing utilization on the ESXi hosts. VAAI is enabled by default in the ESXi hosts and supported by almost all modern All-Flash Array. Always ensure that VAAI is enabled.
7. Enable Space Reclamation on Datastores and Guest-OS Due to the small capacity footprint and the expensive it could be, All-Flash Array gains a tremendous benefit when space reclamation is enabled. Space reclamation allows to thin-provisioned environments removing the dead space from datastore and virtual disks; this dead space is caused by files that have been removed or moved. In VMware environments, always enable space reclamation at the VMFS datastore and virtual machine level.
Juan Mulford is a Solutions Architect at AccelStor. He has held different roles in IT for over 9 years, working with Satellite Communications, Networking, IP Surveillance, Storage and Virtualization technologies. For the last 6 years, he focuses on VDI solutions. He aims to research, design and test AccelStor’s terrific AFA system in virtual environments, supporting the Marketing and Sales Operations at AccelStor.