Abstract

Temperature Sensitive Storage (TSS) is a data storage architecture that keeps track of the access frequencies of data and uses this information to manage storage efficiently and effectively. The term temperature refers to the access frequency of the data.

TSS is a hierarchical storage architecture with frequently accessed data residing in a storage device with faster access time, while less frequently accessed data placed in a device which is more storage efficient. The hierarchy is currently a 3-tier architecture consisting of declustered RAID 1 (RAID1), RAID 5 (RAID5) and compressed RAID 5 (cRAID5) storage devices. The data is dynamically moved between these tiers, depending on their access frequencies.

TSS, along with fast access and efficiency of storage, also provides reliability of data by incorporating redundancy in the system through the use of RAID architecture.

Simple policies for dynamically managing the data have been implemented as user level daemon processes. They are independent of the low level implementation of the TSS driver. They make use of certain ioctl() calls provided by the TSS device driver.

In this paper we discuss the design, implementation and policy framework of a Linux based TSS, and compare its performance with static RAID systems¹.