• Keith Browning (12/1/2010)


    ...or you could use perfmon to get the same.

    True, but there's some overhead with constantly running a perfmon. Just scraping the already collected data from the DMV's is less overhead. We do something similar to this with the Memory Grants Pending metric. That's usually an indicator of memory saturation (ie we're dead in the water), but by itself is not that useful because it's not a predictive metric. It only tells me my server is dead, it doesn't tell me anything prior to the server becoming saturated.

    Page Life Expectancy is a decent predictive metric. It's not the end-all-be-all metric for memory consumption, but at least it'll start an investigation which is usually all we need.