Citation Context

... first. The d-Choices algorithm has been studied extensively in a balls-and-bins, hashing and load balancing setting (e.g., [2,12,15]) and was also proposed as a GC algorithm for solidstate drives in =-=[10]-=-, a paper that is being published concurrently. The latter paper also proposes a mean field model for uniform workloads, but the system operation seems to differ significantly from ours, as the write ...

Citation Context

... fewest number of valid pages. • The Windowed GC algorithm [11], which tries to combine the simplicity of the FIFO algorithm with the performance of the Greedy algorithm. • The d-choices GC algorithm =-=[20, 13]-=-, which selects the block with the fewest number of valid pages out of a set of d randomly chosen blocks. Although the Greedy algorithm minimizes the write amplification in case of uniform random writ...

Citation Context

...cent papers empirically study the intrinsic characteristics of SSDs (e.g., [1], [5]), or develop analytical models for the write performance (e.g., [9], [15]) and garbage collection algorithms (e.g., =-=[23]-=-) of SSDs. Bit error rates of SSDs are known to increase with the number of erase cycles [12], [27]. To improve reliability, prior studies propose to adopt RAID for SSDs at the device level [2], [16],...

Citation Context

...by the notion of “cleaning” a block. Then the particle process models an infinite SSD memory cell with k-page blocks, and with file sizes equal to the page size. The remainder of the paper stays with the natural SSD terminology as it keeps the application in view at no cost. SSD performance studies are an increasingly important, fast growing research area. Space constraints prevent us MAMA 2013 Pittsburg, PA USA Copyright is held by author/owner(s). from covering the literature in full; for a much more comprehensive coverage of the literature, we refer the reader to two, just published papers [1, 3]. Each of these papers studies interesting garbage-collection optimization problems, and both apply mean-field approximation techniques. Garbage collection refers to the cleaning of blocks having a relatively large fraction of inactive files (garbage), a process that must be preceded by copying to clean pages in other blocks the relatively few still-active files. Garbage collection reclaims clean storage space, and makes for more efficient utilization of storage, but it comes with non-trivial costs: Cleaning times are relatively long, and blocks have finite lifetimes in the sense that there is...

Citation Context

...[19, 14, 25, 5] selects the blocks in a cyclic order. • The Greedy GC algorithm [2, 5] selects the block containing the fewest number of valid pages among all the blocks. • The d-choices GC algorithm =-=[21, 13, 22]-=- selects the block with the fewest number of valid pages out of a set of d randomly chosen blocks. • The Windowed GC algorithm [8] maintains a window containing the w least recently selected blocks an...

Citation Context

...ion on DiskSim with SSD add-ons, we demonstrate how RGA can be parameterized to realize the performance–durability tradeoff. An earlier conference version of the paper appeared in ACM SIGMETRICS 2013 =-=[36]-=-. In this journal version, we present all proofs, revise the presentation of mean-field model, analyze write amplification, and provide more rigorous arguments for our discussion throughout the paper....

Citation Context

... d-Choices algorithm has been studied extensively in a classic balls and bins, hashing and load balancing setting (e.g., [2, 15, 18]) and was also proposed as a GC algorithm for solid-state drives in =-=[12]-=-, a paper that is being published concurrently. The latter paper also proposes a mean field model for uniform workloads, but the system operation differs significantly from ours, as the write operatio...