The only supported devices for an IQ multiplex, multi-node configuration, is raw devices. That makes it quite simple for multiplex.
For simplex, you have a choice. However, I've found that using raw devices is faster. With filesystems we must disable caching and use direct IO. This has to be done to guarantee data integrity. As a general rule, filesystems using direct IO are slower than raw devices. I know, I know this filesystem and that OS is faster. Yes, there can be cases where that is true. But as a general rule across all platforms, I see better performance out of raw devices.
Most filesystems tend to take the nice large chunks of IO that IQ does and break them into tiny pieces thus destroying quite a lot of our performance. Since caching is disabled, we lose that little performance trick, too. We also lose control of object and device placement with filesystems. We lose all device isolation that IQ allows with raw devices. That is unless you want to build a filesystem for each LUN. But that defeats the purpose of filesystems.
Plus adding devices is faster with raw devices as we don't need to initialize anything. We simply read the device size, take it all, add it to IQ, and start using it. With filesystems we must build the file device which can take some time if you are trying to build terabyte size devices.
Mark
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.