Scanrail - Fotolia
When the data center world embraced virtualization, the move to shared storage followed. As time progressed, the expense of that storage continued to grow and organizations started to look at different ways to reduce the cost of I/O.
Hyper-converged infrastructure and local disk caching have become hot products designed to reduce storage costs. These products take advantage of the local server's ability to access local attached inexpensive disk or solid-state drives (SSDs). One of the challenges, however, is that these local disk caching approaches offered by companies, such as PernixData, or hyper-converged infrastructure appliances from the likes of Nutanix can cost hundreds of thousands of dollars in software and hardware. Even software-based approaches, such as VMware's Virtual SAN still come with a hefty license cost and some hardware requirements.
Local storage has been on a roller coaster ride as administrators and analysts go back and forth over whether it deserves a spot in the virtualized data center. This tug of war means we see a lot of empty drive slots in a traditional rack server, while we boot our hypervisor with a USB stick. Believe it or not, those empty drive bays can still serve a purpose and without high costs or third-party software.
When you decide to take advantage of these empty slots, the key to success is deciding what type of data or VMs to host there. VMs and data on local drives should not be critical production data, nor does it have to be throw-away data. Think of it more like data you can afford to lose but would rather not. While this sounds cryptic, let me explain and the possible uses will jump out.
In any environment, test servers are created and destroyed for a variety of reasons. They fall into a few categories and are easy to identify:
- The "just testing an application" server -- With many new applications, IT wants to try before it buys. This may require multiple VMs. Why should this application sandbox go on shared storage? If an application moves to production, it should be rebuilt to keep it as clean as possible. With IT trying so many different applications -- only a small collection of which will become production applications -- why start them on shared storage if they will go nowhere?
- The "one-time upgrade" server -- There are always some applications that only need a test and development platform during that once-a-year upgrade. These are often clones of running production servers with a life span of a few weeks. These make ideal platforms to run locally and not affect your shared storage frames.
- The "I just need a few TBs until my disk comes in" server -- If I had a nickel for every time I heard that one. We all know application folks develop and deploy first before getting the needed infrastructure. They have a deadline to meet, even if it means you have to scramble. Why not give them what they want right away? Of course, you'll have to let them know that the performance might not be exactly what they are used to until the new disk arrives. Later, a few storage vMotions, and everyone is happy.
- The "IT VDI or tools" servers -- Unfortunately, IT never seems to be able to buy shared storage for itself, but is still asked to support infrastructure 24x7, from anywhere. Having dedicated VDI sessions for IT personnel gives them the ability to leave the laptop behind. This approach even allows for virtual servers to be loaded with older versions of management consoles and tools that modern operating systems no longer support. These virtual servers or desktops can become critical when doing after hours support.
Is SSD the answer to local disk selection? Well, not exclusively, but a combination of capacity and speed gives you the greatest flexibility based on your planned workloads. Of course, with anything good there is a downside. If the local host goes offline, you no longer have access to these VMs. These VMs will not restart on another host. The key with this idea is selecting VMs that can afford to be offline for a while as they are restored from backups. While this may not give you back all of your storage, it can help and act as a buffer for your organization to be as flexible as possible without breaking the bank.
Will local storage for virtualization catch on?
How local storage solves some problems but creates others
Improving local storage performance on virtual servers
Local storage vs. cloud storage architecture
The impact of local storage for VDI