You Can’t Do Performance Analysis with SMI-S and Other Myths

Some vendors are perpetuating the myth that SMI-S is not designed for performance management. Recently some of our customers asked a vendor to surface additional performance metrics through SMI-S. They received a response along the lines of: “SMI-S is not supposed to handle performance metrics; it is mainly for management. If you want performance metrics, you should buy our proprietary tool.”

While SMI-S has some limitations, the SMI-S Block Server Performance (BSP) defines a very rich set of storage system components for which metrics can be defined: System, Peer Controller, Front-end Adapter, Front-end Port, Back-end Adapter, Back-end Port, Replication Adapter, Volume, and Disks. The BSP further defines counters for reads, writes, read throughput, write throughput, read response time and write response time for each of the components. Coupled with the comprehensive configuration information available within the standard, the SMI-S standard provides a rich canvas for a client consuming SMI-S data to paint the performance profile of a vendor’s hardware.

The BSP sub-profile specifies the components and metrics required for minimal BSP support – unfortunately, not a very long list. I have recently proposed an advanced version that adds the following requirement to the minimal implementation:

Select two of the following three components and provide latency:

1. Front-end Port
2. Volume
3. Disk (or substitute by utilization)

Front-end port and volume measurements provide the analyst with insight into whether or not a performance issue is related to front-end components, and the disk measurements provide the analyst with the information necessary to understand if the issue is on the back end.
The BSP Advanced option provides an optional implementation for vendors with the goal of rewarding those that have already enriched their provider beyond the standard, which ultimately moves the standard forward.

While it is true that an implementation that just conforms to the minimal required set within the BSP would be insufficient for conducting proper performance management, the SMI-S BSP provides plenty of room for optional support. Kudos to EMC, IBM, and HP for providing way more than minimal conformance and giving performance analysts everywhere the support to conduct performance analysis using SMI-S.

While I can’t blame vendors for trying to sell their proprietary tools, I can blame them for obfuscating and for not providing other methods to get the performance metrics out of their systems.

So I ask, who is to blame for insufficient performance metrics: the vendor, or SMI-S?