Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

The nitty-gritty differences between SAN and NAS architectures

I hear the terms that NAS is file-oriented storage while SAN is block oriented. Yet NAS can share files by various locking schemes so I'm not sure as to the advantage or disadvantage of block versus file storage architectures. Besides the effects of NAS requests going up and down the IP stack, are there really any performance penalties in a Gigabit Ethernet versus a Fibre Channel SAN? Do you have a good white paper or reference that can explain, at a low level, the difference between the two architectures?


Network Attached Storage provides data as a remote file system. Basically, a redirector on the server where the application is executing turns a file request for data into a remote file system access via TCP/IP over the local area network. The NAS device handles the network request and turns that into direct block I/O for the data. So, even though accessing data over the network is a file operation, the NAS device turns it into a block I/O operation.

File accesses over the network were the original protocols developed for networks by implementations of NFS and CIFS the most commonly used remote file system access software. Block I/O is what a device understands, so at some point the I/O must be turned into a block access such as a SCSI command (over a SCSI bus, Fibre Channel, etc.). The performance issues are: the introduced latency due to this conversion; traversing the TCP/IP protocol stack; the physical connection; and competition for the resource. (Bad things happen when the network gets congested). The actual bit rate of the interface is a very misleading number. For example, Gigabit Ethernet should transfer data at 100 MB/s but the average transfer rate is much less than that because of various overhead along the way. Fibre Channel is typically 95% efficient due to the fact that it is a hardware intensive implementation (at a cost obviously) with special information passed to do things such as pre-allocating hardware buffers, etc.

We teach a seminar where we go into details about the whys and wherefores of the differences but have not specifically written a paper on it. I have not seen an objective paper on the differences. Unfortunately, vendor papers are trying to sell their product and that should always be taken into consideration.

Randy Kerns

Dig Deeper on Primary storage devices

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.