In LSM trees, background compaction tasks contend with foreground queries for CPU cycles, cache space, and also SAN bandwidth, if deployed in a disaggregated storage architecture. This study proposesNear-Data Compaction(NDC), which executes compaction on the storage node to utilize its underutilized computing resources. However, enabling NDC introduces several challenges. First, it has to support concurrent file access from both compute and storage nodes. In addition, it has to decide which compaction tasks to be executed on the storage node since the computing resources of a storage node are not unlimited. This study presentsTetherDB, an LSM tree for disaggregated storage architecture that addresses these challenges through lightweight dual-node coordination and selective NDC admission policies. Our evaluation demonstrates that TetherDB improves throughput by up to 2.1× compared to RocksDB in write-heavy workloads.

To explore the opportunity of expanding the page cache using disaggregated memory for file-backed pages, this study presents BalloonStasher, an RDMA-based disaggregated memory for data-intensive applications. Utilizing the ephemeral nature of the page cache, BalloonStasher dynamically adapts to the changing page cache demands of multiple clients. BalloonStasher supports one-sided RDMA-based memory pooling or two-sided RDMA-based memory sharing when using memory nodes. Additionally, it also supports peer memory mode, which utilizes the idle memory of peer nodes. Our extensive performance study compares the benefits and limitations of the three cache modes, and shows that BalloonStasher can mitigate the memory underutilization problem and improve the performance of data-intensive applications by a large margin.

In this work, we explore opportunities to design shared storage systems that leverage the distance connectivity of NVMe over Fabrics (NVMe-oF). NVMe-oF enables the use of NVMe storage devices in a shared storage environment, where multiple servers can access the same storage device via RDMA. Leveraging the distance connectivity of NVMe-oF, we develop a shared file system called EXT4-oF by extending the EXT4 file system. EXT4-oF uses RDMA to enable a local file system to function as a shared file system without requiring remote daemon processes. EXT4-oF employs a novel NVMe-driven lazy cache coherence to maintain cache coherence of file system metadata across multiple compute nodes upon creating new files, all achieved without the need for any daemon processes. To ensure cache coherence, NVMe-driven lazy cache coherence mechanism requires compute nodes to perform a re-read of NVMe-oF to avoid false negative file open errors. Through our experiments, we demonstrate that EXT4-oF improves the performance of MinIO by minimizing network traffic between compute and storage nodes and eliminating the need for TCP/IP communication during remote reads.

As advances in Next-Generation Sequencing have made genome sequence data generation faster and cheaper, the acceleration of genome sequence mapping to the reference genome becomes an increasingly important problem. Much effort has been made to improve the performance of the sequence mapping process.

In this extended abstract, we discuss how to improve access to HDFS by employing NVMe over Fabrics (NVMeoF), which has emerged as a new communication protocol between a host and a storage system. To address the legacy shortcomings of HDFS, we explore the opportunity of enabling all-to-all connections between NVMe SSDs and DataNodes rather than dedicating each NVMe to a single DataNode. Our experimental study shows that we can achieve up to 2.55 times higher I/O throughput than legacy HDFS by making HDFS leverage NVMeoF for remote data access.