Pulling memory usage metrics can cost $4/mo./instance on AWS. What opinions do people have on this cost being worth it on large fleets of EC2? Have you often changed instance type based on memory usage?

Hey John! 

This is a great question and in my experience, it depends on what your workloads are most constrained by. (CPU vs. Memory).  I would personally like to see memory metrics captured at a point when the application workload is normal/stable to better understand how my provisioned capacity is being consumed.  Once we better understand our true consumption, we can make more informed decisions on appropriate instance family / types.  This may be an exercise that isn't done ongoing if your environment isn't changing much, so you can disable / enable the metric as needed to optimize spend.

On the other side of the coin, what is the cost/value (risk) of not enabling custom memory metrics and being under/overprovisioned?  

Pulling memory usage metrics can cost $4/mo./instance on AWS. What opinions do people have on this cost being worth it on large fleets of EC2? Have you often changed instance type based on memory usage?

Thanks Dustin. Good advice.

Hey John! 

This is a great question and in my experience, it depends on what your workloads are most constrained by. (CPU vs. Memory).  I would personally like to see memory metrics captured at a point when the application workload is normal/stable to better understand how my provisioned capacity is being consumed.  Once we better understand our true consumption, we can make more informed decisions on appropriate instance family / types.  This may be an exercise that isn't done ongoing if your environment isn't changing much, so you can disable / enable the metric as needed to optimize spend.

On the other side of the coin, what is the cost/value (risk) of not enabling custom memory metrics and being under/overprovisioned?  

Pulling memory usage metrics can cost $4/mo./instance on AWS. What opinions do people have on this cost being worth it on large fleets of EC2? Have you often changed instance type based on memory usage?

Thanks Dustin. Good advice.

Hey John! 

This is a great question and in my experience, it depends on what your workloads are most constrained by. (CPU vs. Memory).  I would personally like to see memory metrics captured at a point when the application workload is normal/stable to better understand how my provisioned capacity is being consumed.  Once we better understand our true consumption, we can make more informed decisions on appropriate instance family / types.  This may be an exercise that isn't done ongoing if your environment isn't changing much, so you can disable / enable the metric as needed to optimize spend.

On the other side of the coin, what is the cost/value (risk) of not enabling custom memory metrics and being under/overprovisioned?  

Pulling memory usage metrics can cost $4/mo./instance on AWS. What opinions do people have on this cost being worth it on large fleets of EC2? Have you often changed instance type based on memory usage?

Hi all, just wanted to add that there are two additional data sources from which Cloudability is able to pull utilization data for Compute Rightsizing- Datadog and New Relic. With these higher precision data sources, the rightsizing engine surfaces more informed recommendations that can provide additional savings on top of the existing savings opportunities, further reducing cloud spend. As you can imagine, having memory metrics allows Cloudability to generate more accurate recommendations with more confidence. 

Thanks Dustin. Good advice.

Hey John! 

This is a great question and in my experience, it depends on what your workloads are most constrained by. (CPU vs. Memory).  I would personally like to see memory metrics captured at a point when the application workload is normal/stable to better understand how my provisioned capacity is being consumed.  Once we better understand our true consumption, we can make more informed decisions on appropriate instance family / types.  This may be an exercise that isn't done ongoing if your environment isn't changing much, so you can disable / enable the metric as needed to optimize spend.

On the other side of the coin, what is the cost/value (risk) of not enabling custom memory metrics and being under/overprovisioned?  

Pulling memory usage metrics can cost $4/mo./instance on AWS. What opinions do people have on this cost being worth it on large fleets of EC2? Have you often changed instance type based on memory usage?