Title :
Toward millions of file system IOPS on low-cost, commodity hardware
Author :
Da Zheng ; Burns, Randal ; Szalay, Alexander S.
Author_Institution :
Dept. of Comput. Sci., Johns Hopkins Univ., Baltimore, MD, USA
Abstract :
We describe a storage system that removes I/O bottlenecks to achieve more than one million IOPS based on a user-space file abstraction for arrays of commodity SSDs. The file abstraction refactors I/O scheduling and placement for extreme parallelism and non-uniform memory and I/O. The system includes a set-associative, parallel page cache in the user space. We redesign page caching to eliminate CPU overhead and lock-contention in non-uniform memory architecture machines. We evaluate our design on a 32 core NUMA machine with four, eight-core processors. Experiments show that our design delivers 1.23 million 512-byte read IOPS. The page cache realizes the scalable IOPS of Linux asynchronous I/O (AIO) and increases user-perceived I/O performance linearly with cache hit rates. The parallel, set-associative cache matches the cache hit rates of the global Linux page cache under real workloads.
Keywords :
Linux; cache storage; input-output programs; 32 core NUMA machine; AIO; CPU overhead; IO bottlenecks; IO scheduling; Linux asynchronous IO; cache hit rates; commodity SSD; extreme parallelism; file abstraction; file system IOPS; global Linux page cache; lock-contention; low-cost commodity hardware; nonuniform memory architecture machines; parallel page cache; storage system; user space; user-perceived IO performance; user-space file abstraction; Hardware; Instruction sets; Kernel; Linux; Message systems; Performance evaluation; Data-intensive computing; low cost; millions of IOPS; page cache optimization; solid-state storage devices;
Conference_Titel :
High Performance Computing, Networking, Storage and Analysis (SC), 2013 International Conference for
Print_ISBN :
978-1-4503-2378-9
DOI :
10.1145/2503210.2503225