Remediating hotspotting in high-performance compute workloads with ONTAP FlexCache volumes
Suggest changes
PDFs
A common problem with many high-performance compute workloads, such as animation rendering or EDA, is hotspotting. Hotspotting is a situation that occurs when a specific part of the cluster or network experiences a significantly higher load compared to other areas, leading to performance bottlenecks and reduced overall efficiency due to excessive data traffic concentrated in that location. For example, a file, or multiple files, is in high demand for the job running which results in a bottleneck at the CPU used to service requests (via a volume affinity) to that file. FlexCache can help alleviate this bottleneck, but it must be set up properly.
This documentation explains how to set up FlexCache to remediate hotspotting.
|
Beginning July 2024, content from technical reports previously published as PDFs has been integrated with ONTAP product documentation. This ONTAP hotspot remediation technical report content is net new as of the date of its publication and no earlier format was ever produced. |
Key concepts
When planning hotspot remediation, it's important to understand these essential concepts.
-
High-density FlexCache (HDF): A FlexCache that is condensed to span as few nodes as the cache capacity requirements allow
-
HDF Array (HDFA): A group of HDFs that are caches of the same origin, distributed across the cluster
-
Inter-SVM HDFA: One HDF from the HDFA per server virtual machine (SVM)
-
Intra-SVM HDFA: All HDFs in the HDFA in one SVM
-
East-west traffic: Cluster backend traffic generated from indirect data access