Cloud Migrations: Solid Foundation or Shifting Quicksand? Part 2

Cloud Migrations 2Not All Workloads are Created Equal

In the first part of this blog, I explained how cloud operators can assess performance by running a diverse array of tests that focus on specific capabilities, such as CPU, storage, memory and of course, network performance. Spirent CloudScore application performs these tests, providing performance and overall scores for each category. This is extremely valuable for seeing “the big picture” on cloud performance, as well as useful drill-down information on specific areas where performance might be affected after an update to any cloud component, such as a hypervisor, NIC or NIC driver.

Part two of this blog continues to address the cloud update use case, but with a focus on the particular workload for a particular cloud. As mentioned, each organization is unique, and likewise the workload that each organization’s cloud must bear, is also unique. The impact on cloud performance due to any changes to said cloud is very much workload dependent.  In the Netflix example mentioned in the previous blog, although the patches for Meltdown and Spectre introduced a huge potential performance hit of as much as 800% for Netflix’s particular workload, the actual performance impact was only approximately 0.1%. The conclusion is that cloud performance must be verified using the intended workload.

Recording and Replaying Cloud Workloads

VM ActivityTo address this specific use case, Spirent is introducing an exciting new feature for its CloudStress application, the ability to record and replay workloads. This new feature allows cloud operators to record a workload on a cloud by capturing the original workload’s meta-data. In other words, while the specific memory read/write operations are not recorded, the quantity and nature of operations per second are recorded. 

The recorded meta-data workload can then be played back against the original cloud to establish a baseline level of performance. After the cloud has been updated in some way, the recorded meta-data workload can then be played again, this time to assess the performance of the modified cloud. In doing so, CloudStress can assess the performance impact of cloud updates for a specific workload.

How to Adjust Cloud Resourcing

As mentioned, the question is not so much about blaming any part of the cloud infrastructure. After all, if the patches for Spectre and Meltdown are required for security purposes, there is no way to avoid deploying them. However, knowing how to adjust cloud resourcing, considering these changes, streamlines the process greatly. 

If the performance of any group of resources has been reduced, an important question emerges. How much more of that resource type needs to be added to maintain performance levels? Fortunately, CloudStress offers convenient capacity methodologies to answer this specific question. Capacity methodologies help determine resource levels for CPUs, memory, storage, networking, or a mix of these resource types.

Learn how Spirent CloudStress can address your cloud and virtualized infrastructure testing needs.

Read our new white paper on the implications of the Spectre and Meltdown vulnerabilities.


comments powered by Disqus
× Spirent.com uses cookies to enhance and streamline your experience. By continuing to browse our site, you are agreeing to the use of cookies.