It’s a confusing storage world out there. Making the choice about what storage platform you should use (never mind which vendor) is a tricky business. The industry doesn’t make things any easier by keeping the acronyms so similar. We have DAS (Direct Attached Storage), NAS (Network Attached Storage) and SAN (Storage Area Networks). What do these wonderful TLAs mean and why do we have so many storage technology choices? Let’s look at them in turn.
Direct Attached Storage (DAS) It does exactly what it says on the tin – it is attached direct to the server and was the original way disks and hosts interacted. For small-scale installations this is probably still the best choice. The disk is directly available to the host for presentation as file systems. In its favour, DAS is cheap and easy to install and support. On the negative side, DAS doesn’t scale well and basic installations will not come with the data redundancy of more large-scale products. But, when all you’re paying for is a box of disks, you can’t really expect much else.
It is possible to improve on DAS and use hardware RAID SCSI cards to build in some redundancy, or fibre channel connected disks to get more physical scalability from your solution. However, once you start down that route you might as well consider one of the larger scale solutions - NAS or SAN.
Network Attached Storage (NAS)
It's hard to believe that someone didn’t think to make the acronyms for NAS and SAN a bit more different: people are forever confusing the two. Network Attached Storage (NAS) is storage accessed over an IP network using either the NFS or CIFS protocols (NFS is a Unix protocol, CIFS is a Windows protocol). Both of these protocols effectively operate on data at the file level, so when you make a change, you commit the whole file. This is great for the user in marketing who is editing a spreadsheet that gets changed infrequently (as the spreadsheet is in memory on the local PC), but not so clever for large-scale databases which may have files of hundreds of megabytes in size. For these large scale requirements, bring in SAN, the big daddy of storage infrastructures.
Storage Area Networks (SAN)
Storage Area Networks allow lots of storage units and hosts to be interconnected with an infrastructure known as a fabric, usually drawn as a nice fluffy cloud in SAN diagrams. The fabric consists of multiple switches and fibre-optic cabling, in a similar way to the implementation of IP networks. SAN configurations, however, tend to be more rigidly structured with designs that implement redundancy and minimise the number of switch connections between the host and the disk. Unfortunately, SAN implementations are the most expensive infrastructure to deploy and consequently the most expensive to support. On the plus side, SANs do scale extremely well and if implemented correctly can be very flexible.
So we know the basics, how do we choose which one to implement? Start by taking a look at your requirements. Firstly, are you delivering storage to end users using PC workstations or to servers? If it is the former, then NAS is likely to be your most cost effective solution. The latest NAS servers now include features specifically tailoring the operating system to meet the requirements of workstation-based users, including space quotas, fast reboots for minimal server downtime, low cost replication and snapshots of data for instant backup/restore.
It you’re looking to provide storage to a server then the choice is likely to be DAS or SAN. Both infrastructures provide disks, which appear to the host to be SCSI attached local disks (DAS because they are; SAN because the host bus adaptor (HBA) in the server simulates SCSI protocols) and therefore provide block level access to files. As a rule of thumb, I tend to ask myself the following questions:
- Do I need more than 10-20GB of storage?
- Will this ever increase?
- Am I happy with a simple hardware or host-based RAID solution?
At this point it is worth mentioning hybrid technologies like iSCSI and SAN blades. These are blurring the boundary between NAS and SAN. iSCSI provides SCSI protocols across IP networks, either using a dedicated hardware adaptor or software driver. This seems on the face of it a cheaper solution than implementing a dedicated SAN. However, most administrators may choose to implement dedicated iSCSI networks for performance reasons, rather than having iSCSI traffic traverse the production network, making the cost choice one of IP switches over fibre channel switches. SAN blades allow NAS boxes to deliver disks into a Storage Area Network. These have definite potential where reducing the cost of the disk subsystem is of key importance, for example in a development environment.
Whatever your disk requirements, the options for delivering storage continue to grow and become more complex. We can guarantee the decision on storage platforms is not going to become simpler in the coming years.