Amazon's latest 'flex' VMs promise savings for your burstiest apps

Sustained workloads need not apply

Updated Amazon Web Services added another set of cost-optimized instances to its EC2 lineup on Tuesday, aimed at customers whose workloads aren't pegging the CPU 100 percent of the time.

The cut-rate C7i-flex instances are based on the same 4th-gen Intel Xeon scalable processors as its standard C7i VMs. These chips are capable of hitting all-core turbos of 3.2GHz and max turbo clocks of up to 3.8GHz, under ideal circumstances. The key difference is that the flex variant targets bursty workloads with generally low to average utilizations, while the more expensive non-flex VMs are more appropriate for sustained loads.

The cloud giant argues its flex instances – which also extend to its general purpose M7i family – are ideal for those running things like web apps, as the compute demand is likely to fluctuate with traffic throughout the day. In fact, AWS emphasizes – as it would – that these VMs are appropriate for the vast majority of modern applications including databases, caches, Apache Kafka, and Elasticsearch.

AWS's C7i-flex service is available in a number of configurations ranging from two to 32 vCPUs and 4GB to 64GB of DDR5 in a 2:1 ratio of memory to compute. Storage is available via Amazon's 10Gbps Elastic Block Storage service, while networking on each of these instances tops out at 12.5Gb/sec of throughput.

We'll note that customers that need larger VMs or faster storage or networking will need to opt for its full-fat C7i SKUs.

While the benefit of these flex instances to customers is lower pricing, we suspect Amazon is the real winner here. Most cloud infrastructure is shared and, to varying degrees, over provisioned. This means multiple customers may have workloads running on the same server, if not the same cores.

By encouraging customers to move burstier workloads to its so-called flex instances, Amazon is presumably able to over provision to a greater degree, as the likelihood of multiple customers demanding full performance simultaneously is relatively low.

This is probably why Amazon specifically recommends that customers running workloads like batch processing, analytics, or high performance computing – which will peg the CPU for extended periods – use its standard C7i instances instead.

What happens if AWS detects you're running a sustained workload on a flex instance isn't clear. We've asked Amazon for comment regarding what we can only imagine is a fairly common problem and will let you know what we find out.

If we had to guess, because the C7i flex and non-flex instances are based on the same hardware, it could end up getting live migrated to a more expensive SKU if average utilization ends up being too high. ®

Updated to add

Amazon tells us that customers running sustained workloads on its C7i instances will not see their workloads migrated but "might see a gradual reduction in the maximum burst CPU throughput."

It recommends users take advantage of its CloudWatch metrics to monitor their CPU utilization to ensure their workloads are deployed appropriately and consider a shift to non-Flex instances for heavy processing.

More about

TIP US OFF

Send us news


Other stories you might like