Storage and raid disks

Page 1

1 RAID disks The aim is to emulate a very expensive disks buying a bunch of less expensive ones: to increase the performance, the size and the reliability of storage systems, several independent disks are considered as a single one. Data are striped across the disk and accessed in parallel, thus gaining high parallelism in accessing them and load balancing. Two techniques must be implemented on RAID disks. 1. Data striping: data are written sequentially according to a cyclic algorithm (round robin), it allows reading and writing in parallel. a. Stripe unit: the amount of data written on a single disk, b. Stripe width: the number of disks considered by the algorithm, which can be less than the total amount of disks. 2. Redundancy: it is necessary as the probability of failure rises with the growing of the number of disks, it consists of error-correcting codes, that are stored on different disks than the data and can be used to reconstruct the data after a failure; the main drawback is the slowing down of the writing operations, as these values must be computed and stored along with the data. There are many types of architectures, the choice should be done accordingly to the required features. 1. RAID 0 Data are written on a single logical disk and then split by the controller among the several physical disks. It has the lowest costs and the best write performance of all the levels, but the failure of a single disk will cause the loss of data, so it is used where performance and capacity are very important while data reliability is not an issue. 2. RAID 1 Data are duplicated, two physical disks contain the same data; it has a high reliability but read and write performance are not bad, as data can be accessed in parallel without the ned of computing parity bits; the main…