I worked an analysis of the Linux EXT3 file system because at my current employer, we had not put Linux in a tier one SAN based storage environment. While we have done some large applications with a large amount of throughput with linux, we have not used SAN as a storage delivery system with Linux. The NAS implementation was my configuration with tuning I have also posted on this site.
This study was to get some perspective on the different backend storage alternatives. EMC DMX800, Network Appliance 920 filer NAS and fiber SAN connected as well as a Network Appliance 960 filer NAS and SAN connected.
I did not use the latest greatest storage but what I had available in the lab and in truth the results varied more on the amount of memory allocated to cache than the performance of the storage device. There is a lot of data and a lot to interpret from it, but I will leave that to the reader for now.
I used the 4 Linux IO schedulers for further perspective and well I was curious as to their impacts. All using RHEL 4 update 5. I later did some tests with RHEL5 and did not like the results so I have to do some more tuning and analysis. Despite the network driver bug in RHEL4U5, the NAS tests are consistent performance wise. That bug seems to only affect stability in rare occurrences and not throughput. Data measured using iozone in a directio configuration. Tuning parameters are shown in the last page.
Click Here => Linux IO Scheduler Document