Non-volatile memory (NVM) is a new storage technology that combines the performance and byte addressability of DRAM with persistence traditional devices like flash (SSD). While these properties make NVM highly promising, it not yet clear how to best integrate into layer modern database systems. Two system designs have been proposed. The first use exclusively, i.e., store all data index structures on it. However, because has higher latency than DRAM, this design can be less efficient...

A hybrid storage system consisting of SSDs and HDDs improves the performance user IOs by moving IO concentrated areas from to SSDs. It is important for identify such move them immediately. We studied concentrations operational servers found one type that they continue up an hour within a narrow region volume. The either same continued area or which moves neighboring after 3 minutes in average had passed. Moreover, it appeared at random logical block addresses (LBAs). decide call Wandering...

Hybrid storage systems consisting of SSDs and HDDs improve the performance input-output (IO) access by migrating IO concentrations from to SSDs. Therefore, it is important for system identify such immediately migrate them This study analyzes logs shared filesystems, Microsoft (MS) Exchange Server, Virtual Desktop Infrastructure (VDI) viewpoint locality. illustrates common features locality portions volume elapsed time. Moreover, we examine selection a hybrid these features. We have...

This paper proposes a scalable communication layer for multi-dimensional hyper crossbar network using multiple Gigabit Ethernet the PACS-CS system which consists of 2560 single-processor nodes and 16 x 10 three dimensional hyper-crossbar (3D-HXB). To realize high performance existing networks, host processor usage processing must be reduced to less than appropriate packet time is calculated from message size target bandwidth. overcome this problem, we have developed PM/Ethernet-HXB facility....

The response time of solid state drives (SSDs) has dramatically reduced according to the spread non-volatile memory express (NVMe) devices. These devices have times less than 100 micro seconds on average. all-flash-array systems also drastically through use NVMe SSDs. However, there are applications, particularly, virtual desktop infrastructure and in-memory database systems, that require storage with even shorter time. Their workloads were found contain many input-output (IO)...

In a previous workload study, we found new type of I/O concentration approximately 6 GB in size that continues for more than 10 minutes and then moves on to another area. We also the conventional hybrid storage method, static tiering caching, is not sufficient this concentration. The on-the-fly automated (OTF-AST) designed cannot handle other concentrations well, so have developed system composed both OTF-AST caching. proposed system, dynamically migrates SSD caching handles concentrations....

Various vendors develop a byte accessible Nonvolatile Dual-Inline Memory Module (NVDIMM). The performance of the NVDIMM drastically surpasses that Solid State Drive (SSD), which is connected by PCI express. However, cost much higher than SSD. Therefore, hybrid storage system between and SSD an effective technique for improving cost-performance. If uses less while maintaining performance, its cost-performance should be improved. Our previous work involves on-the-fly automated tiering...

The response times of solid state drives (SSDs) have decreased dramatically due to the growing use non-volatile memory express (NVMe) devices. Such devices less than 100 micro seconds on average. all-flash-array systems also through NVMe SSDs. However, there are applications, particularly virtual desktop infrastructure and in-memory database systems, that require storage with even shorter times. Their workloads tend contain many input-output (IO) concentrations, which aggregations IO...

Automated tiered storage with fast memory and slow flash (ATSMF) is a hybrid system located between non-volatile memories (NVMs) solid state drives (SSDs). ATSMF aims to reduce average response time for inputoutput (IO) accesses by migrating concentrated IO access areas from SSD NVM. However, the current implementation cannot sufficiently because of bottleneck caused Linux brd driver, which used NVM driver. The driver more than ten times larger speed. To sufficiently, we developed...

Emerging solid state drives (SSDs) based on a next-generation memory technology have been recently released in market. In this work, we call them low-latency SSDs because the device latency of is an order magnitude lower than that conventional NAND flash SSDs. Although can drastically reduce I/O perceived by application, overhead OS processing included has become noticeable very low latency. Since executed CPU core, its operating frequency should be maximized for reducing overhead. However,...

Hybrid storage techniques are useful methods to improve the cost performance for input-output (IO) intensive workloads. These choose areas of concentrated IO accesses and migrate them an upper tier extract as much possible through greater use areas. Automated tiered with fast memory slow flash (ATSMF) is a hybrid system situated between non-volatile memories (NVMs) solid-state drives (SSDs). ATSMF aims reduce average response time by migrating access from SSD NVM. When finishes, migrates...

In previous studies, we determined that workloads often contain many input-output (IO) concentrations. Such concentrations are aggregations of IO accesses. They appear in narrow regions a storage volume and continue for durations up to about an hour. These occupy small percentage the logical unit number capacity, include most accesses, at unpredictable block addresses. We investigated these by focusing on page-level regularity found they few regularities. This means simple caching may not...

Automated tiered storage with fast memory and slow flash (ATSMF) is a hybrid system located between non-volatile memories (NVMs) solid state drives (SSDs). ATSMF reduces the average response time for input-output (IO) accesses by migrating concentrated IO-access areas from an SSD to NVM. However, current implementation cannot sufficiently reduce because of bottleneck caused Linux brd driver, which used NVM access driver. The driver more than ten times longer memory-access speed. To time, we...