iTransformer: Using SSD to Improve Disk Scheduling for High-performance I/O

Abstract

Abstract-The parallel data accesses inherent to large-scale data-intensive scientific computing requires that data servers handle very high I/O concurrency. Concurrent requests from different processes or programs to hard disk can cause disk head thrashing between different disk regions, resulting in unacceptably low I/O performance. Current storage systems either rely on the disk scheduler at each data server, or use SSD as storage, to minimize this negative performance impact. However, the ability of the scheduler to alleviate this problem by scheduling requests in memory is limited by concerns such as long disk access times, and potential loss of dirty data with system failure. Meanwhile, SSD is too expensive to be widely used as the major storage device in the HPC environment.We propose iTransformer, a scheme that employs a small SSD to schedule requests for the data on disk. Being less space constrained than with more expensive DRAM, iTransformer can buffer larger amounts of dirty data before writing it back to the disk, or prefetch a larger volume of data in a batch into the SSD. In both cases high disk efficiency can be maintained even for concurrent requests. Furthermore, the scheme allows the scheduling of requests in the background to hide the cost of random disk access behind serving process requests. Finally, as a non-volatile memory, concerns about quantity of dirty data are obviated. We have implemented iTransformer in the Linux kernel and tested it on a large cluster running PVFS2. Our experiments show that iTransformer can improve the I/O throughput of the cluster by 35% on average for MPI/IO benchmarks of various data access patterns.

iBridge: Improving Unaligned Parallel File Access with Solid-State Drives

“…Accesses that are well aligned in the logical file address space can be mapped to unaligned disk accesses because of data striping and using I/O optimization techniques such as collective I/O and data sieving [26] in MPI-IO middleware. Data prefetching techniques [6], [8], which hide I/O time behind computation time, become less effective as a program can spends more time on I/O than on computation, and with high access concurrency due to uncoordinated requests from different processes of an MPI program [31]. Zhang et al proposed a data-driven execution mode to improve I/O efficiency via program pre-execution when program performance hinges on I/O resources [33].…”

“…iBridge gives priority to more performance-critical unaligned data access than regular random requests. iTransformer [31] was recently proposed to help disk schedulers handle high I/O concurrency. Requested data are either committed to disks or to SSDs according to the I/O workload's spatial locality.…”

A Checkpoint of Research on Parallel I/O for High-Performance Computing

“…Their design focuses on wear-awareness, response time and hit ratio. This ideaa fast SSD as a cache for an HDD -is also explored by Zhang, Davis, and Jiang [117].…”

HyCache: A User-Level Caching Middleware for Distributed File Systems

“…A more general multi-tiering scheme was proposed in [21] which helps decide the needed numbers of SSD/HDDs and manage the data shift between SSDs and HDDs by adding a 'pseudo device driver', again, in the kernel. iTransformer [22] considers the SSD as a traditional transient cache in which case data needs to be written to the spinning hard disk at some point once the data is modified in the SSD. iBridge [23] leverages SSD to serve request fragments and bridge the performance gap between serving fragments and serving large sub-requests.…”