Managing IOPS

Best Practices for Managing IOPS in your Cloud

Every data center must plan for Input and Output Operations Per Second (IOPS) in their environments. Each type of storage (SAN) can handle various limits of IOPS capacity. Some of these limits are expansive and can handle a lot of sustained traffic while others cannot. Depending on your traffic patterns, IOPS can create problems if you ignore sound management principles. Understanding how to optimize IOPS is critical to the success of using cloud services not only for you, but your customers as well. 

For example, a blog server would have low levels of IOPS, but a high transactional database could potentially use your entire IOPS capacity at peak times, which also vary. Peak times may be early morning or steady throughout the day, but more likely and most often IOPS traffic peaks during backups that are done overnight.

Any time you are doing simultaneous backups in a cloud environment you may be utilizing multiple paths without being aware of it. This can create additional traffic and hinder the flow of data, or in a worse case event, high traffic can cause the data flow to stop. Transaction logs can fill up with errors and stop processing requests. A server can begin to use SWAP space and create further issues. Because the cloud is trying to clean up from unsound management practices, it returns slow response times, creating unhappy customers.

Before you begin utilizing and reselling cloud, these are some things to consider when you look at how your cloud will be utilized by your employees and your customers.

  1. Active databases are usually accessed hundreds or thousands of times per second, creating heavy traffic. If this is a typical use of your environment, you should plan to utilize Premium SAN for this type of storage.
  2. Staggering backups help minimize high and sustained peaks of IOPS, and will help your utilization of cloud services to flow smoothly.
  3. Overlapping backups of the same virtual machine (VM) can slow the system or freeze the VM to protect the other VMs in the cloud. If you need multiple backups of the same server be sure to stagger them so that one is finished prior to the start of the next.
  4. Monitor your IOPS traffic and logs in your network operation center (NOC). This is a must for every environment, but especially in a cloud where services are provisioned on demand or can be rapidly scaled. Remember to look at the reports that include your entire cloud, not just a particular customer account.
  5. If a server begins to use SWAP, increase the RAM for that VM to eliminate SWAP usage, as it leads to unneeded IOPS traffic volumes.

In general, Premium SAN can burst up to 30K and Primary SAN can handle peak bursting between 8K and 10K. If you have sustained traffic at these levels rather than bursts, you may see performance issues that linger past your backup schedule into the morning, interfering with business. Monitoring and planning ahead before you reach capacity IOPS is the best method to keep the cloud running smoothly and your customers happy with their cloud service.

photo credit: A dome of cloud