I recently caught an episode of a documentary series my kids were watching on MTV, called “Catfish.” In the show, people who were in online-only relationships tried to find out if the person they’d been chatting with was actually who she/he claimed to be. It won’t come as a surprise to anyone that they usually weren’t. Think the Manti Te’o hoax.
As someone in the IT world, I began to think about other ways in which the virtual and physical IT worlds don’t always line up. As we turn to virtualization and the Cloud, we often struggle to integrate physical devices to support virtual systems. And just like in “Catfish,” the reality is often not what we’d hoped.
This is particularly true for SANs. We’re going from a one server, one application world to one in which many systems all access the same storage. The performance of traditional SANs has often been inadequate, and the difference between locally attached and virtually provisioned storage can be very noticeable.
Why is that? What aspects of traditional SANs are in conflict with virtual infrastructures? The fundamental problem is a lack of what I call “VM awareness.”
Physical vs. Virtual System Performance (… Real People vs. Facebook Profiles)
A traditional iSCSI SAN uses LUNs as the basic management abstraction. This presents problems when they are integrated with a virtual architecture, which is managed using VMs and virtual disks. The volumes are configured and managed independent of the virtual infrastructure, which requires administrators to manage mapping and create VM policies to effectively manage the storage.
Beyond the complexity of managing a SAN for a virtual environment, traditional SANs also suffer from latency and random disk-read issues, because they are designed to handle predictable input and output patterns from one system. Over the years, predictive algorithms have been developed to anticipate upcoming operations, so they can be called up in anticipation of being needed. (This is called prefetching.) This improves the performance of the SAN and reduces wait times.
But with virtual architectures, this whole pattern is turned on its head. Now, multiple servers and machines are all accessing the storage at the same time in a much more random pattern. (This is often called the I/O Blender.) This can easily overwhelm the system, and cause performance bottlenecks that make the system lag or even freeze.
Bridging the Physical and Virtual (… Know Who You’re Talking To)
A SAN needs to integrate several features in order to effectively accommodate a virtual environment.
- First, it needs to be able to handle a large number of IOPS. The IOPS (input-output per second) of traditional hard drives won’t cut it, even high-speed SATA drives. Solid state drives (SSDs) provide superior read/write speeds, but performance tends to degrade with large writes, and they have a higher per-GB cost. A hybrid storage design gives the best of both worlds. SATA drives can be used for long-term storage of inactive data, while an SSD cache handles active reads and writes.
- Second, it needs to be able to prioritize data so that the most frequently accessed data is always at hand, while other data is stored away (aka – storage tiering). This requires that you distinguish between active and inactive data. Manually tiering data is a hassle and requires ongoing attention, so a SAN needs to automatically tier data to serve a virtual infrastructure conveniently.
- Third, the system needs to deduplicate data blocks. Otherwise, the same blocks will be stored multiple times, which slows down performance and fills up the storage rapidly. SSDs are more expensive than hard disks, so you want to be sure that cache capacity is maximized through block-level deduplication.
- Lastly, storage management needs to integrate with VM management. Ideally, a single interface can be used to manage the provisioning of both your virtual machines and storage. Your SAN should be managed in conjunction with your virtual systems, rather than be forced to work with them. Look for a solution that provides both storage and virtualization control in one, and also accommodates business continuity, since this impacts data growth substantially.
In the end, the most important takeaway here is that you need a fundamental awareness of how virtual infrastructures differ from physical ones, and be sure that the technology is designed with this understanding in mind. If the episode of “Catfish” taught me anything, it’s that you need to validate in order to ensure compatibility.
Zenith Infotech specializes in delivering enterprise-class IT solutions for virtualization, storage, and backup & recovery. Zenith partners with IT providers serving the SMB market so they can capitalize on the lucrative cloud market. Zenith’s TigerCloud virtual infrastructure is built from the cloud down to provide a complete converged infrastructure. Learn more at www.zenithinfotech.com or email [email protected].