Hi, I'm an Oracle DBA for the dutch government. In private I'm an enthousiastic btrfs user for some (2) years and I'm looking forward to introducing it at work as as RedHat will support it (hopely with RHEL7). Last couple of weeks i've been testing different storage options for Oracle Database files: - LVMStriping with ext4 - Software Raid with LVM and ext4 - Software Raid directly with ext4 (without LVM) - btrfs (directly on partitions) - ASM To optimize performance I've been actively looking at some IOstat numbers. Btrfs did not do really good and I think I know the main reason. Both with ext4 and ASM Oracle could do large 'multiple block writes' and large 'multiple block reads' (looking at avgrq-sz). With btrfs I found that the latency was actually better, but the avgrq-sz was verry small. So, much IO's but small IO's and therefore not so much MBPS. I think that this is because the 'The stripe size is fixed to 64KB for now. '... Please be aware that this is the 'technology preview version of btrfs with RH6'. This is not exactly bleeding edge, so some performance enhancements yet comitted might not be part of this tests. However, this got me thinking. Thing is that I had to set a stripesize for both Software Raid and LVM Striping. ASM is a little optimized in this part, being that it can set a stripe size per filetype (datafiles 1MiB and redo 128k, as I recall). So large IO's can be written with large requests and small IO's can be striped accross many disks. If there is a possibility to let the stripesize be dependent on the request size for a write, you might have simmilar to as for ASM optimisation, and better, sinds it is not depending on 'the filesystem knowing the filetype'. I think that this would be a great enhancement for large files with big IO's. And might give btrfs a lead to more conventional filesystems like ext4 and xfs. Is there an option to bring this in development? Is it allready thought of, planned? When could it be available? And if practical I could get involved somehow. Perhaps testing with workloads. Please let me know on your thoughts on this matter... -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html