Customer request: Split OS and data disk storage partitions to improve disk performance
Executive Summary:
During the investigation of a service outage reported by a customer (Parkland), a review of Engage VM disk layout was performed by the customer. The customer observed the VM deployment uses a single logical volume and VMDK file to store both OS and data. The customer technical team requested the OS and data be stored in separate partitions as a best practice for I/O performance.
Background Information:
Between late February and early March 2019 Parkland experienced some intermittent Engage service outages as a result of an unexpected VM migration to a lower-grade hardware host compared to the original host. The VM migration was not immediately identified by Parkland’s technical team for about two weeks, eventually when the VM was migrated out of the problem-host the performance returned to expected level and no further outages were reported. During the two week interim there were a number of emails and live discussions with the Parkland team about possible causes and resolution to the outages. Parkland requested a temporary CLI account in order to monitor the system during the times of outages and also support the migration efforts. With access to the system they made observations on the disk layout. Currently all Engage 5.X VM deployments uses a single logical volume and VMDK file to store both OS and data. The Parkland team requested that the OS and data be stored in separate partitions for as a best practice for performance improvements.
Historically splitting the OS from the application data has been a method to improve performance, i.e. individual disk transactions are serviced by two different devices offers a possible performance improvement by avoiding bottlenecks when the system requires simultaneous OS and application access. However, in modern virtualized systems with smart cacheing, multiple gigabytes of RAM and large-scale datastores there’s no substantive data to show a split of the OS and data into separate partitions would improve performance.
Attached document: Performance Best Practices for VMware vSphere 6.7 (Chapter 2 ESXi Storage Considerations)
Additional Documentation:
There is a Performance Best Practices for VMware vSphere 6.7 (Chapter 2 ESXi Storage Considerations) pdf that I could not include in the idea. I can provide that at request.
During the investigation of a service outage reported by a customer (Parkland), a review of Engage VM disk layout was performed by the customer. The customer observed the VM deployment uses a single logical volume and VMDK file to store both OS and data. The customer technical team requested the OS and data be stored in separate partitions as a best practice for I/O performance.
Background Information:
Between late February and early March 2019 Parkland experienced some intermittent Engage service outages as a result of an unexpected VM migration to a lower-grade hardware host compared to the original host. The VM migration was not immediately identified by Parkland’s technical team for about two weeks, eventually when the VM was migrated out of the problem-host the performance returned to expected level and no further outages were reported. During the two week interim there were a number of emails and live discussions with the Parkland team about possible causes and resolution to the outages. Parkland requested a temporary CLI account in order to monitor the system during the times of outages and also support the migration efforts. With access to the system they made observations on the disk layout. Currently all Engage 5.X VM deployments uses a single logical volume and VMDK file to store both OS and data. The Parkland team requested that the OS and data be stored in separate partitions for as a best practice for performance improvements.
Historically splitting the OS from the application data has been a method to improve performance, i.e. individual disk transactions are serviced by two different devices offers a possible performance improvement by avoiding bottlenecks when the system requires simultaneous OS and application access. However, in modern virtualized systems with smart cacheing, multiple gigabytes of RAM and large-scale datastores there’s no substantive data to show a split of the OS and data into separate partitions would improve performance.
Attached document: Performance Best Practices for VMware vSphere 6.7 (Chapter 2 ESXi Storage Considerations)
Additional Documentation:
There is a Performance Best Practices for VMware vSphere 6.7 (Chapter 2 ESXi Storage Considerations) pdf that I could not include in the idea. I can provide that at request.
-
Guest
- Jan 31 2020
- Will not do
-
- +1
Related ideas
