Heterogeneous disk arrays are becoming (or will be in a near future) a common configuration in many sites. Let us describe two scenarios that end up in a heterogeneous disk array. The first one appears whenever a component of a traditional array fails and it has to be replaced by a new one. As disk technology improves quite rapidly, it is quite probable for the new disk to be faster and larger than the ones already in the array [7]. A similar scenario appears when the capacity needs of a site grow and new disks have to be acquired to grow the size of the array (by increasing the number of disks in the array). In this case, it will also be difficult to buy the same disks as the ones in the original configuration, and thus newer disks will be added. In both cases, we will make the array a heterogeneous one because it will be made of disks with different characteristics. This kind of situation is especially common in low-cost clusters of workstations, where cost is an important issue and old components have to be used as well as possible. According to the study performed by Dr. Grochowski at IBM [7], disk capacity nearly doubles every year while the price per Mbyte is decreasing about 40% per year. This means that the price of arrays will remain about the same throughout the years, although the capacity will be increased a lot, of course. If a given site wants to buy a 32 disk array (assuming for example 18GB Seagate disks at today prices), it costs between $17000 and $26400 (depending on the interface, RPM, and seek time) [18]. At this price, changing all these disks at a time because one of them breaks is too expensive for many institutions and/or companies, especially if the problem can be solved by just buying a single disk. The only exception appears when the site does not need to grow its capacity and thus replacing the 32-disk array by a few new ones (reducing the size of the array) is enough. Nevertheless, this does not seem to be the trend as disk usage grows constantly.
To handle this kind of disk array, current systems do not take into account the differences between the disks. All disks are treated as if they had same capacity (the smallest one) and performance (the slowest one). This is not the best approach because improvements in both capacity and response time of the heterogeneous array could be achieved if each disk were used accordingly to its characteristics.
In this work, we present a simple solution to this problem by proposing AdaptRaid5, a block-distribution algorithm that improves the performance and effective capacity of heterogeneous disk arrays compared to current solutions. We should note that this proposal has been especially evaluated for scientific and general purpose workloads (understanding as workload the requests that reach the disk controller, after being filtered by the-file system cache) because the multimedia case has already been addressed quite successfully by other research groups [6, 17, 24]. Nevertheless, the proposed algorithm also works well in a multimedia environment.
This paper is divided into 8 Sections. Section 2 presents the most relevant work in the area of heterogeneous disk arrays. Section 3 introduces the reader to some important concepts that need to be clarified before describing the algorithm, which is explained in full detail in Section 4. Section 5 presents the methodology used to obtain the results presented in Section 6. Section 7 presents the future work we plan to do in this field. Finally, Sections 8 and 9 present the conclusions that can be extracted from this work and how to get more information about this work.