3 ms·
Think about it in a cloud model where you aren’t renting a server you know by name. Imagine running something like object storage or FaaS where the cloud provid
by acdha 8y ago
Think about it in a cloud model where you aren’t renting a server you know by name. Imagine running something like object storage or FaaS where the cloud provider handles everything behind the scenes and can failover at any time without you seeing more than perhaps a few failed requests.
In that model, hardware failing is just a factor in total overhead cost. If the hardware doesn’t fail immediately it might be cheaper to leave a dead node in the rack than to pay a human to touch it, especially if they’ve already recouped a significant percentage of the purchase cost by the time it fails. Over the life of a server the cost of cooling is enough that a substantial savings will push that breakeven point earlier.