Global namespace acceleration

Global namespace acceleration (GNA) enables data on node pools without SSDs to have additional metadata mirrors on SSDs elsewhere in the cluster. Metadata mirrors on SSDs can improve file system performance by accelerating metadata read operations.

You can enable GNA only if 20 percent or more of the nodes in the cluster contain at least one SSD and 1.5 percent or more of total cluster storage is SSD-based. For best results, before enabling GNA, make sure that at least 2.0 percent of total cluster storage is SSD-based.

Even when enabled, GNA becomes inactive if the ratio of SSDs to HDDs falls below the 1.5 percent threshold, or if the percentage of nodes containing at least one SSD falls below 20 percent. GNA is reactivated when those requirements are met again. While GNA is inactive in such cases, existing SSD mirrors are readable, but newly written metadata does not get the extra SSD mirror.

Note Image

Node pools with L3 cache enabled are effectively invisible for GNA purposes. All ratio calculations for GNA are done exclusively for node pools without L3 cache enabled. So, for example, if you have six node pools on your cluster, and three of them have L3 cache enabled, GNA is applied only to the three remaining node pools without L3 cache enabled. On node pools with L3 cache enabled, metadata does not need an additional GNA mirror, because metadata access is already accelerated by L3 cache.