Storage That Only Looks Like A SAN

In my last column, I discussed how some vendors are abandoning shared storage for virtualized environments in favor of local storage. Their goal is to reduce cost and complexity while increasing performance.

Through the use of inter-host mirroring and replication they can still provide many of the key features of virtualization, but there are some problems: You need a complete second copy of a virtual machine (VM) on another host, you are limited to only that second host for failover or migration (unless you make multiple copies), and there is CPU consumption required of the second target VM. Essentially, you double your VM count and the resources those VMs require. In a resource-constrained environment, this could be a problem.

Vendors are trying to deliver other solutions that keep the cost, simplicity, and performance advantages of local storage solutions but that still provide VM flexibility and efficiency. One approach is the SAN-Less SAN.

[ For more on shared vs. local storage, see Is Shared Storage's Price Premium Worth It? ]

The SAN-Less SAN is actually another form of shared storage, but the storage is in the physical hosts of the virtual infrastructure instead of on a dedicated shared storage system. Each host is equipped with hard drives or Flash SSD storage, and as data is being stored it is written across each host in the infrastructure--similar to how data is written across the nodes of a scale-out storage cluster.

Redundancy is achieved by using a RAID-like data stripping technique so that failure of one host or the drive of one host does not crash the entire infrastructure. As in traditional RAID, the redundancy is provided without requiring a full second copy of data. Also, it is not uncommon for the disks in each node to themselves be RAIDed via a RAID card inside the server.

This technique of striping data across physical hosts provides the VM flexibility. All the hosts can get to the VM images, so a VM can be migrated in real time to any host.

One downside of the SAN-Less SAN approach is that you lose the performance advantage of pure local storage since parts of the data must be pulled from the other hosts. From a performance perspective, you have essentially created a SAN.

As discussed in my article, Building The SAN-Less Data Center, some vendors are merging features of local storage with this SAN-Less technique to bring the best of both worlds. These vendors are keeping a copy of each VM data local to the host on which it is installed in addition to replicating the VM’s data across the host nodes. The value of this technique is that the VM gets local performance until it needs to be migrated. A second step in migration allows the newly migrated VM to have its data rebuilt on its new host, restoring performance. This is especially intriguing if the local data is PCIe Solid State Disk.

Of course, nothing is perfect, and the network that interconnects these hosts must be well designed. There is also some host resource consumption as the software that runs the data replication on each host does its work. However, that consumption should not be as significant as a host loaded down with target VMs in the mirroring/replication example discussed in my last column. Finally, the type of hard disks and solid state disks used in the hosts in a SAN-Less SAN must also be carefully considered.

Despite the advantages of local storage and SAN-Less SANs, shared storage is far from dead. In my next column, I will look at local storage vs. SANs.

Even small IT shops can now afford thin provisioning, performance acceleration, replication, and other features to boost utilization and improve disaster recovery. Also in the new, all-digital Store More special issue of InformationWeek SMB: Don't be fooled by the Oracle's recent Xsigo buy. (Free registration required.)