I/O considerations

MSDP works best with multiple file systems configured to provide the disk space to store its data and metadata. Ideally, each file system should be created on independent storage volumes with equal size and no disk/LUN sharing for best parallel I/O operations. If possible, MSDP metadata should be configured to be stored on a different file system separate from the file systems storing the data containers due to its different I/O patterns.

The size of a file system configured for an MSDP pool may be in the range of several tens of TB up to 100 TB. For each file system, MSDP has dedicated worker threads and data buffers for data writes, compactions, and CRC checks, etc. If the size of the file systems for data containers vary a lot, the smaller file systems may be filled up earlier than the larger ones. The smaller file systems stop receiving data when filled up which reduces the I/O bandwidth and eventually affects server performance. The impact will be more significant, especially for I/O intensive workloads, such as low deduplication ratio backups, optimized duplication, or restores, the last two are read-intensive as the data needed is not likely found in file system cache.