Feature to improve write performance with Thin and Lazy Zeroed Thick VMFS based virtual disks in ESXi 7.0U2
Article ID: 335037
Updated On:
Products
VMware vSphere ESXi
Issue/Introduction
There are three virtual disk types supported by VMFS. Namely Thin, LZT (Lazy Zeroed Thick) and EZT (Eager Zeroed Thick).
The use cases of these virtual disks are different. However there is substantial difference in IO performance among these virtual disks. The first-write performance of Thin and LZT virtual disks is very low compared to the EZT virtual disks.
The telemetry data shows Thin and LZT virtual disks types are widely deployed compared to EZT, with Thin being the predominant type of virtual disk.
The Thin disks are predominantly used for space efficiency. For Thin virtual disks, blocks are allocated and zeroed on demand and for LZT virtual disks blocks are allocated upfront, but zeroed on demand. Such "first writes" incur the cost of zeroing and a transaction to update on-disk metadata, which results in low performance. This feature aims at specifically improving the “first write” performance with "Thin" and "Lazy Zeroed Thick" virtual disks. This will be useful for customers who want to use Thin disk (take advantage of space efficiency) without paying penalty for performance.
The use cases of these virtual disks are different. However there is substantial difference in IO performance among these virtual disks. The first-write performance of Thin and LZT virtual disks is very low compared to the EZT virtual disks.
The telemetry data shows Thin and LZT virtual disks types are widely deployed compared to EZT, with Thin being the predominant type of virtual disk.
The Thin disks are predominantly used for space efficiency. For Thin virtual disks, blocks are allocated and zeroed on demand and for LZT virtual disks blocks are allocated upfront, but zeroed on demand. Such "first writes" incur the cost of zeroing and a transaction to update on-disk metadata, which results in low performance. This feature aims at specifically improving the “first write” performance with "Thin" and "Lazy Zeroed Thick" virtual disks. This will be useful for customers who want to use Thin disk (take advantage of space efficiency) without paying penalty for performance.
Environment
VMware vSphere ESXi 7.0.0
Resolution
As described, the feature will improve write performance for both Thin and LZT virtual disks. We expect performance improvement and reduced turnaround time in many work flows involving these virtual disks.
For example:
- We expect significant performance improvement while writing or backing up data on a Thin disk. The increased Io performance will also reduce overall back up time for customers who use Thin or LZT disks to back up data.
- We also expect similar improvement in IO performance and reduction in restore time while restoring backed up data to a Thin or LZT Disk.
- For Thin disks supporting unmap or space reclamation, there is always opportunity of “first write” and the feature improves overall IO performance of such work flows. For example we can expect performance improvement for work flow involving storing current and deleting old surveillance videos.
- We also expect improvement in work flow such as SRM (Site Recovery manager) or VR (vSphere Replication ) when the target disk is Thin or LZT type.
Please note: The quantum of improvement depends on type of underlying storage ( all-flash, magnetic disk , NVMe etc), configuration and the exact work flow used by the customer. The benefits may vary for each customer, however in general we expect improved IO performance for Thin and LZT disks.
For example:
- We expect significant performance improvement while writing or backing up data on a Thin disk. The increased Io performance will also reduce overall back up time for customers who use Thin or LZT disks to back up data.
- We also expect similar improvement in IO performance and reduction in restore time while restoring backed up data to a Thin or LZT Disk.
- For Thin disks supporting unmap or space reclamation, there is always opportunity of “first write” and the feature improves overall IO performance of such work flows. For example we can expect performance improvement for work flow involving storing current and deleting old surveillance videos.
- We also expect improvement in work flow such as SRM (Site Recovery manager) or VR (vSphere Replication ) when the target disk is Thin or LZT type.
Please note: The quantum of improvement depends on type of underlying storage ( all-flash, magnetic disk , NVMe etc), configuration and the exact work flow used by the customer. The benefits may vary for each customer, however in general we expect improved IO performance for Thin and LZT disks.
Feedback
Yes
No
Powered by
