We introduce the scheduler subversion problem, where lock usage patterns determine which thread runs, thereby subverting CPU scheduling goals. To mitigate this problem, we introduce Scheduler-Cooperative Locks (SCLs), a new family of locking primitives that controls lock usage and thus aligns with system-wide scheduling goals; our initial work focuses on proportional share schedulers. Unlike existing locks, SCLs provide an equal (or proportional) time window called lock opportunity within which each thread can acquire the lock. We design and implement three different scheduler-cooperative locks that work well with proportional-share schedulers: a user-level mutex lock (u-SCL), a reader-writer lock (RW-SCL), and a simplified kernel implementation (k-SCL). We demonstrate the effectiveness of SCLs in two user-space applications (UpScaleDB and KyotoCabinet) and the Linux kernel. In all three cases, regardless of lock usage patterns, SCLs ensure that each thread receives proportional lock allocations that match those of the CPU scheduler. Using microbenchmarks, we show that SCLs are efficient and achieve high performance with minimal overhead under extreme workloads.

We introduce Sky, an extension to the VMM that gathers insights and information by intercepting system calls made by guest applications. We show how Sky gains three specific insights -- guest file-size information, metadata-data distinction, and file-content hints -- and uses said information to enhance virtualized-storage performance. By caching small files and metadata with higher priority, Sky reduces the runtime by 2.3 to 8.8 times for certain workloads. Sky also achieves 4.5 to 18.7 times reduction in the runtime of an open-source block-layer deduplication system by exploiting hints about file contents. Sky works underneath both Linux and FreeBSD guests, as well as under a range of file systems, thus enabling portable and general VMM-level optimization underneath a wide range of storage stacks.

Benchmarking file and storage systems on large file-system images is important, but difficult and often infeasible. Typically, running benchmarks on such large disk setups is a frequent source of frustration for file-system evaluators; the scale alone acts as a strong deterrent against using larger, albeit realistic, benchmarks. To address this problem, we develop David: a system that makes it practical to run large benchmarks using modest amount of storage or memory capacities readily available on most computers. David creates a “compressed” version of the original file-system image by omitting all file data and laying out metadata more efficiently; an online storage model determines the runtime of the benchmark workload on the original uncompressed image. David works under any file system, as demonstrated in this article with ext3 and btrfs. We find that David reduces storage requirements by orders of magnitude; David is able to emulate a 1-TB target workload using only an 80 GB available disk, while still modeling the actual runtime accurately. David can also emulate newer or faster devices, for example, we show how David can effectively emulate a multidisk RAID using a limited amount of memory.