Researching System Administration - Anderson (Correct)
....both reads and writes the mean request rate, request size, queue length and sequentiality, as described in the analysis section. The input queue length and sequentiality are adjusted to take into account the effect of interactions between streams on the same LU using the techniques described in [UAM01] The sequentiality is decreased for two streams that are on simultaneously, because the overlap will cause extra seeks. The queue length is increased because there will be more outstanding I Os, giving the disk array more opportunity for reordering. The utilization of each stream is calculated ....

Mustafa Uysal, Guillermo A. Alvarez, and Arif Merchant. A Modular, Analytical Throughput Model for Modern Disk Arrays. In To appear in Proceedings of the 9th International Symposium on Modeling, Analysis and Simulation on Computer and Telecommunications Systems (MASCOTS 2001), Cincinnati, OH, August 2001.

--------------------------------------------------------------------------------
Ergastulum: Quickly Finding Near-Optimal Storage.. - Anderson.. (1 citation) (Correct)
....supports a given workload. Analytic models are better than trace driven simulators because they are faster. Ergastulum treats the performance models as a black box, making frequent calls to them through a narrow interface. Ergastulum is currently used with several different performance models [30, 3, 21], depending on which device types are used. The storage system design problem is complex. With capacity only constraints and fixed size disks, it is as hard as the NP complete bin packing problem [13] For arbitrary goals or models only an exhaustive search would find an optimal design. Consider ....

....the workload, summarizes an I O request pattern. Streams capture information about the I O accesses to a single associated store, such as an average request rate, an average request size, and a summary of the sequentiality in block accesses. Further information on stream attributes can be found in [30, 3, 21]. Disks in an array are grouped together into logical units (LUs) using RAID [23] Stores are packed into the LUs of configured disk arrays. The array types available to the design tool are specified as inputs, along with any restrictions on their configuration. Disk arrays inherently have ....

[Article contains additional citation context not shown here]

M. Uysal, G. A. Alvarez, and A. Merchant. A Modular, Analytical Throughput Model for Modern Disk Arrays. In Proceedings of the 9th International Symposium on Modeling, Analysis and Simulation on Computer and Telecommunications Systems (MASCOTS 2001.

--------------------------------------------------------------------------------
CHAMELEON: a self-evolving, fully-adaptive resource arbitrator - For Storage Systems Self-citation (Alvarez) (Correct)
No context found.

M. Uysal, G. A. Alvarez, and A. Merchant. A modular, analytical throughput model for modern disk arrays. In Proc. of the 9th Intl. Symp. on Modeling, Analysis and Simulation on Computer and Telecommunications Systems, pages 183--192, August 2001.

--------------------------------------------------------------------------------
Hippodrome: Running Circles Around Storage Administration - Anderson, Hobbs, Keeton.. (2002) (39 citations) Self-citation (Uysal) (Correct)
....This section describes the components of the Hippodrome loop in more detail, and explains how the components interact to design a storage system iteratively. The Hippodrome components are the most recent versions of our group s ongoing research into storage system modeling and configuration [2, 3, 4, 5, 6, 10, 26, 30, 31, 32]. The goal of this paper is to show how the different components can work together to automate storage management. Therefore, this section summarizes the techniques used in Hippodrome. The details on each component may be found in the paper on that topic. Hippodrome uses four interdependent ....

....number of candidate configurations. The performance models have three complementary parts: 1. Inter stream adjustments. The input queue length and sequentiality are first adjusted to take into account the effect of interactions between streams on the same LU using the techniques described in [30]. For example, the sequentiality is decreased for two streams that are on simultaneously, because the overlap will cause extra seeks, while the queue length is increased because there will be more outstanding I Os, which gives the disk array more opportunity for request re ordering to improve ....

M. Uysal, G. Alvarez, and A. Merchant. A modular, analytical throughput model for modern disk arrays. In 9th International Symposium on Modeling, Analysis and Simulation on Computer and Telecommunications Systems (MASCOTS

--------------------------------------------------------------------------------
Selecting RAID levels for disk arrays - Anderson, Swaminathan, Veitch, A. (2002) (11 citations) Self-citation (Alvarez) (Correct)
.... There is no single best choice of RAID level: the right choice for a given store is a function of the access patterns on the store (e.g. reads versus writes; small versus large; sequential versus random) the disk array s characteristics (including optimizations such as write buffer merging [23], segmented caching [26] and parity logging [21] and the effects of other workloads and stores assigned to the same array [18, 27] In the presence of these complexities, system administrators are faced with the tasks of (1) selecting the type and number of arrays; 2) selecting the size and ....

....that workload. In the results reported in this paper, we use our thirdgeneration solver, Ergastulum [5] prior solver generations were called Forum [7] and Minerva [1] Our solvers are constraint based optimization systems that use analytical and interpolation based performance models [3, 7, 18, 23] to determine whether performance constraints are being met by a tentative design. Although such models are less accurate than trace driven simulations, they are much faster, so the solver can rapidly evaluate many potential configurations. As illustrated in Figure 1, the solver designs ....

M. Uysal, G. A. Alvarez, and A. Merchant. A modular, analytical throughput model for modern disk arrays. In International Symposium on Modeling, Analysis and Simulation on Computer and Telecommunications Systems (MASCOTS), August 2001.