Balance Resource Consumption in VMware Cloud on AWS

Imbalanced storage, CPU and memory consumption across your networks can be a costly problem. We’ll analyze an anonymous sample organization, based on real data, to break down these costs. Then we’ll explore a range of solutions, made possible when you’re using VMware Cloud on AWS.

You’ll find a number of  efficiency-enhancing design factors in VMware Cloud on AWS architecture. One of the most significant of these is the ability to balance compute and storage resources across the platform. 

In an ideal world, you’ll always consume and exhaust storage, CPU, and memory resources in parallel.

Figure 1. Linear Scaling and Balance of Compute and Storage Resources

 

Unfortunately, the reality is that organizations often have imbalanced resource consumption. Let’s explore this problem with a sample organization based on anonymized data extracted from RVTools

We’ll call this sample organization ABC Corporation. Let’s say ABC Corporation has entered the number of VMs and the average vCPU, memory, and storage into the VMware Cloud on AWS Sizer and TCO Tool. They find they’ll need 52 nodes on i3.metal hosts to run their production environment (see Figure 2, below).

In VMware Cloud on AWS i3.metal hosts, NVMe storage is local to the host. There’s no way to increase that attached storage (10.7 TB raw capacity). In order to accommodate storage requirements, ABC Corporation will have to pay for additional hosts. They won’t be able to utilize the extra compute resources of these hosts. This imbalanced resource consumption is inefficient and costly.

Figure 2. Example of Unbalanced Resource Consumption

 

Let’s take a closer look at the details in Figure 2. ABC Corporation learns that only 5% of the workloads consume approximately 80% of the storage resources. If they take out that 5% of storage-intensive resources and focus on the remaining 95%, ABC Corporation will need only 12 nodes instead of the 52 nodes previously calculated. So, ABC Corporation will need 11 nodes for  90% and 10 nodes for 80% of the resources.

Fortunately for ABC Corporation, there are 3 storage offerings available in VMware Cloud on AWS(see Figure 3, below): 

1/ Direct-attached NVMe storage 

2/ Elastic vSAN which is a fusion of VMware vSAN and Amazons Elastic Block Store (EBS), 

3/ External storage array

Figure 3. VMware Cloud on AWS Storage Offerings

 

Direct attached NVMe is available in i3.metal hosts (36 Cores, 512 GB RAM, 10.7 TB Storage).

Elastic vSAN is available in R5.metal hosts in VMware Cloud on AWS (48 Cores, 768 GB RAM, 15 – 35 TB EBS Storage). 

The third option, external storage array is on the roadmap and currently only available as a Managed Service Provider (MSP) solution. It comes in the form of Network File System (NFS) datastore. At the time of writing, only two MSPs provide this solution: Faction and Rackspace. (Want to learn more? Watch the VMworld session by Glenn Sizemore and Martin Hosken, ‘VMware Cloud on AWS Storage Architecture and Design Options’.)

Not all workloads are created equal. Some workloads don’t require the high-performance of Non-Volatile Memory Express (NVMe) storage. ABC Corporation has a large amount of inactive (cool) data that falls into this category. For example: video libraries for training material and log data for compliance. 

ABC Corporation can consider external storage options for their cool data. They could make minor changes to application design or data topology. These changes could lead to a significantly reduced node count – and noticeable cost savings.

Figure 4. Media Delivery Application

 

ABC Corporation can also utilize native Amazon storage services such as Amazon Simple Cloud Service (S3), Amazon Elastic File System (EFS), Amazon FSx, and other native Amazon services. Each service has a unique combination of performance, durability, cost, and interface. Figure 4 depicts an example of a hybrid storage architecture for ABC Corporation Media Delivery Application using S3 and EFS.

Note that an external storage solution like EFS and FSx is mounted to guest VMs instead of ESXi hosts. That way it can take advantage of the high bandwidth, low latency ENI connection available in VMware Cloud on AWS to access native AWS services.

Part of the larger workloads in ABC Corporation are databases. Depending on their architecture, internal compliance may require ABC Corporation to leave some of these workloads on-prem. Our sample organization also has the option to migrate data to native AWS services.Amazon Relational Database Service (RDS) for example may help reduce their latency requirements.

In summary, it’s important to examine the workloads in your environment. Seek opportunities to use hybrid storage architecture to achieve a balanced resource consumption and optimize costs.