When it comes to determining storage for virtualization, it is worth a full reconsideration of storage protocols....
Historically, there are two main ways to approach virtualization storage: Fibre Channel SAN and Ethernet-based storage protocols with iSCSI or NFS. Emerging options for virtualization storage will include converged networking adapters (CNAs) that can offer 10 Gigabit Ethernet and Fibre Channel over Ethernet to servers.
The storage protocols variety pack
A Fibre Channel SAN has a relatively high initial investment to create a storage fabric. These storage protocols usually include a pair of switches and HBAs for each server connected to the SAN fabric. Tape drives, disk arrays or SAN storage controllers would also be attached to the SAN fabric. Fibre Channel storage protocols have high performance with speeds of 1, 2, 4 or 8 Gb per second. Fibre channel storage provides multi-pathing support across the two or more switches and the back-end connectivity on storage arrays.
The Ethernet-based storage protocols -- iSCSI and NFS -- can use existing Ethernet infrastructure instead of a dedicated storage networking equipment inventory. In most environments, virtualization host systems will dedicate a number of 1 GB Ethernet interfaces to communicate to the iSCSI or NFS storage. In the case of VMware environments, multiple VMKernel interfaces -- driver stack for NFS and iSCSI storage -- would be allocated to multiple Gigabit Ethernet adapters to provide multipathing for software-based iSCSI connectivity.
Storage protocols for virtualization
When it comes to deciding which storage protocol is right to use for virtualization storage, there is no clear answer. Additional purchase costs and existing investment in storage are usually two critical decision points when it comes to making a decision about storage for virtualization.
Many organizations favor an Ethernet-based storage protocol for virtualization storage because that infrastructure is usually in place and has a lower per-port cost than a Fibre Channel SAN. Many organizations also establish tiers of virtualization storage for various workload classifications. For example, development systems may be placed on less expensive iSCSI storage when top-tier production workloads are placed on Fibre Channel storage.
Virtualization storage and disk selection
When it comes to selecting storage for virtualization, decision makers must also battle with disk types. Most virtualization environments choose between serial attached SCSI (SAS) and serial ATA (SATA) disk types. SAS drives are usually very fast and relatively small compared to their SATA counterparts, but they are also roughly twice as expensive. Cutting edge installations can use solid state disk (SSD) technology for extremely high throughput.
Frequently provisioning storage for virtual environments is a manifestation of what application vendors list in their requirements for an application, which may or may not include virtualization-specific information. In a virtualization storage environment, multiple systems are aggregated to the same disks through various RAID technologies.
Simply speaking, one virtual machine (VM) may touch 50 hard drives in a run state. Likewise, one hard drive may provide part of the I/O for 50 VMs. This means that any application requirements need to be taken with consideration for a consolidated infrastructure. As environments push larger VMs with higher I/O requirements into production, I/O may be the new bottleneck. Once a drive technology is selected, then the same performance arguments can be held on what type of RAID technology to use with overhead taken into account.
Like hypervisors and networks, virtualization storage touches every aspect of a virtualized infrastructure. Planning storage for virtualization should not be taken lightly, and you must pay particular attention to choosing storage protocols. The problem is that there is no one-size-fits-all answer for virtualization storage. What is clear, however, is to carefully weigh options with all factors considered. These include cost, performance and existing investments.
About the author
Rick Vanover (firstname.lastname@example.org), vExpert, VCP, MCITP, MCTS, MCSA, is a virtualization expert in Columbus, Ohio. He is an IT veteran specializing in virtualization, server hardware, operating system support and technology management. Follow Rick on Twitter @RickVanover and click here for Rick’s blogger disclosure.