Modeling and optimizing I/O throughput of multiple disks on a bus

Abstract

In modern I/O architectures, multiple disk drives are attach e d t o e a c h I/O controller. A study of the performance of such a r c hitectures under I/O-intensive w orkloads has revealed a performance impairment that results from a p r eviously unknown form of convoy behavior in disk I/O. In this paper, we describe measurements of the read performance of multiple disks that share a SCSI bus under a heavy workload, and develop and validate formulas that accurately characterize the observed performance (to within 12% on several platforms for I/O sizes in the range 16{128 KB). Two terms in the formula clearly characterize the lost performance seen in our experiments. We describe techniques to deal with the performance impairment, via user-level workarounds that achieve greater overlap of bus transfers with disk seeks, and that increase the percentage of transfers that occur at the full bus bandwidth rather than at the lower bandwidth of a disk head. Experiments show bandwidth improvements of 10{20% when using these user-level techniques, but only in the case of large I/Os.

Access-Time Aware Cache Algorithms

“…3, which is the main contribution of our work, do not depend on a particular structure of the function T (s i ). Nevertheless, we describe here a specific form based on past research on HDD I/O throughput [3] [4], and based on our performance study of disk access time observed in caching servers. We will refer to this specific form later to clarify some properties of the optimal policy.…”

“…A consequence of such decoupling is that, at any time, the RAM stores a subset (M R ) of the contents stored in the HDD (M H ). 3 In our work we consider the same decoupling principle. As a consequence, our policy is agnostic to the replacement policy implemented at the HDD (LRU, FIFO, Random, .…”

On the effect of a configuration choice on the performance of a mirrored storage system

“…The seek minimization properties of physical mirroring with 2 disks with respect to the nearest server algorithm have been studied extensively in [12] and [21]. As we noted there are many other papers which deal with this and related subjects which contain a basic error [5], [6], [7], [9], [10], [26], [25], [28], [29]. Among those whose results remain qualitatively correct despite the error one should note [25] which considers synchronous writes as well as reads and [26] which provides a comparison of online and offline algorithms all assuming physical mirroring.…”

The Locality of Searchable Symmetric Encryption

“…Algorithmic performance with data stored on disk has been studied extensively in external memory models (c.f. [26,23,3]). These models usually consider block-oriented devices with varying degrees of precision (e.g., including modeling parallelism, drive geometry, memory hierarchies, caching, locality of blocks, etc.).…”