More so in virtual desktop environments than with any other application, the demand on server and storage resources can rise sharply due to frequently recurring events that have high I/O demand, such as boot, login, logoff, shutdown, application patches, or running an antivirus scan. We commonly refer to these events as I/O storms. As one of our product managers puts it: it’s just like living in Kansas, getting hit by a tornado twice a day and a hurricane every week. On the storage side of things, we measure and size performance in two ways: (1) data transfer, measured by megabytes per second (MBPS); and (2) number of operations, measured by input/output per second (IOPS) (you can find more on this at the Storage Performance Council website). A VDI environment is typically not that heavy on data transfer, so we care mainly about the number of operations that the hosting infrastructure can support.
A typical IOPS profile of a desktop under normal workload can is 5 to 10 IOPS, depending on the OS and applications the user is using; but the difference in IOPS requirement between regular desktop operations and during one of the events that we listed above could be 10 times higher. Compound that with the number of virtual desktop users consolidated on dense server and storage infrastructure, and you have a big outburst of IOPS requirement that we call the I/O storm – which can be a boot or login storm when users are coming to work in the morning, a logoff or shutdown storm when they leave in the evening, or a combination of both during lunch time. Each one of these events can take anywhere between 30 minutes to 2 hours; and if the environment is not properly sized to deal with these storms, you can have a significantly degraded level of service during these hours.
Jerome Wendt of DCIG describes VDIs as the killer app for storage virtualization, and he outlines all the reasons why storage virtualization is particularly suitable for VDI environments. I’d like to focus more on the economies of scale of the infrastructure and how we size a VDI environment to weather frequent I/O storms.
Mitigating the impact of I/O storms typically requires sizing the environment for the worst case scenario. Traditionally, IOPS are sized based on the number of spindles in the storage environment, so let’s take the example of a company with 5,000 employees. If every employee requires 10 IOPS for normal operations, that would mean that the storage infrastructure would have to deliver 50,000 IOPS to support them. Now if you consider that during morning hours or in the evening there is a maximum number of employees of 10% that are trying to login or log off at the same time, this will create a surge of IOPS requirement that will cause the I/O storm. And let’s assume that the average IOPS requirement for each one of these activities is 5 times the normal operation requirement, this will mean you need to account for 10% x 5 = 50% additional IOPS. This will vary from one organization to the other, but in this example the total required IOPS will be 75,000 IOPS.
If we size the above environment for a (50/50) read/write I/O profile with Fibre Channel or SAS 15,000 rpm drives, you would need around 625 spindles (this will vary according to type of drive and RAID configuration) – this is a massively scalable environment with a lot of wasted capacity to address the IOPS requirements. With FalconStor NSS SAN Accelerator we mix solid state storage with high-density SATA drives to address the IOPS and capacity requirements independently.We size for IOPS with SSD drives or solid state memory arrays that can deliver approximately 5,000 IOPS per drive, which means we can address the 75,000 IOPS requirement with 15 SSD drives while addressing the capacity requirements with SATA drives; this translates into huge reductions in acquisition costs and floor space required, which in turn translates into significant reductions in power and cooling for the data center.
The storage cost of supporting a similar environment as the one in our example would be around $35 per user.
In my next blog, I will address the data protection features that come included with our solutions for VMware View.
We will be demoing the VMware View SAN Acceleration solution in our booth at VMworld 2010 next week. If you are at VMworld this year, come by our booth #1407 to take a peek at the solution, or you can follow us live from VMworld on our FalconStor @ VMworld 2010 website.
