Research Of Hybrid Storage Based On File Access Randomness And Hotness

With the development of storage technology, a variety of new storage devicescontinue showing up. As a new kind of storage device, SSD is far better than HDD onrandom read and write, but it has an exorbitant price. HDD, known as traditional storagedevice, has got a much bigger capacity per disk, but its data access speed has grown quitelittle which has become a outstanding performance bottleneck. Under the circumstance ofthat SSD cannot take the place of traditional HDD with its own cost-performance ratio,hybrid storage will play a very good role in the transition stage. Hybrid storage fullyexploits the advantages of different storage devices to improve the overall performance.How to design migration algorithm and choose migration set according to thecharacteristics of the system and the goodness of different devices is a hot researchproblem in hybrid storage.This paper deeply analysis and compares the performance characteristics of SSD andHDD, concludes what kind of access patterns they each fit better through experiments,designs a new hybrid storage system based on file access randomness and hotness. Bycollecting file access history information, the system abstracts randomness and hotness foreach individual file and comes up with a algorithm to calculate them. The algorithmcomputes a file value with randomness and hotness which means how much benefit thesystem can obtain by migrate that file to SSD. With file value, the system puts forward amigration algorithm to decide the migration set and automatically redirects to theresponding device for the migrated files by using a tool called FUSE at last.The experiments show that after taking both access randomness and hotness intoconsideration, the whole system can gain a performance improvement about10%. And bydistinguishing read and write on SSD and reducing write operations on SSD as much aspossible, SSD can get protection in some way.