Comparing Single Instance VM SLAs of Azure and AWS
Customers often ask us about the SLA between Azure and AWS since they get conflicting claims from sources. The purpose of this blog post is clarify the actual uptime covered by the SLA (as of Nov 2023) for both AWS and Azure for single instance VMs.
This is important especially for SAP deployments where VM instances are over sized and while DR is provisioned in another region (usually), the actual downtime guaranteed by the cloud service provider matters to the customer.
MS Azure:
MS Azure has recently changed their SLA pages and have consolidated them into this URL (good luck wading through the 102 page document!). Virtual machines are in page 86. The key point to be noted is this definition in page 5.
“Applicable Period” means, for the 30 days prior to and including the last day of the Incident for which a Service Credit is owed, the number of days that you are a subscriber for a Service.
Essentially, for single instance VMs (i.e. VMs that are not in AZs and not placed in Availability sets), here is the SLA & associated service credits:
Note:
1) Please note that the highest level of 99.9% is applicable only if ALL the disks attached to the VM are either Premium SSDs or Ultra Disks. If you use multiple disk types, the lowest SLA of all the disks will apply.
Amazon Web Services:
Here’s the SLA on AWS for single instances. The highest level of SLA (‘uptime percentage’) claimed is 99.5% failing which there is a service credit of 10%.
Comparison:
Measuring Downtime: Prima facie, it appears that Azure offers a higher SLA for single instances, viz., 99.9% on Azure vs 99.5% on AWS. However, these SLAs deserve a closer inspection since there are cases where AWS could fare better for customers.
For example, Azure’s ‘Applicable Period’ is defined as 30 days and the downtime is the ‘accumulated minutes’ during that period. So technically, a downtime of up to 43.83 minutes contiguously could still be within the acceptable limit and not breach the SLA. (Downtime (in minutes) = (100% – 99.9%) * (30 days * 24 hours * 60 minutes) ≈ 43.83 minutes per month)
In comparison, AWS’s SLA promises that ‘AWS will not charge you for any Single EC2 Instance that is Unavailable for more than six minutes of a clockhour’. Hence any downtime that crosses 6 minutes contiguously breaches the AWS SLA!
Disk Types and Downtime: While Azure’s SLA for single instances appears to be higher, viz., 99.9%, there are cases where AWS SLA fares better. For example, a VM with Standard HDD attached has only a 95% uptime guarantee on Azure, whereas AWS’s SLA covers it at 99.5%!
This is a good example of how customers can easily have misconceptions about the delivered service based on SLA claims from competing Cloud service providers. (And also why @SaveCloudBills continues to thrive helping customers wade through these questions!)
Takeaways:
1) Despite the apparently better Azure SLA of 99.9% vs AWS’s 99.5%, for workloads that rely on single instance VM use cases (e.g. SAP instances), AWS doesn’t fare so badly since it promises not to have a downtime of more than 6 minutes in a clock hour. Azure’s SLA apparently permits a downtime of up to 43 contiguous minutes without breaching the SLA.
2) While AWS’s SLA is straightforward, Azure’s VM SLA varies by disk types (Premium SSDs /Ultra Disks, Standard SSDs and HDDs). For a VM with standard HDD attached, AWS’s single instance VM SLA fares better than the Azure SLA.
Posted by Gokul – CloudRunr.
