In this paper, we propose BDL-APT (Block Device Layer with Automatic Parallelism Tuning) that maximizes the throughput of IP-SAN protocols in long-fat networks. BDL-APT parallelizes data transfer using multiple IP-SAN sessions at a block device layer, and adjusts the number of active IP-SAN sessions automatically according to network status. A block device layer is a layer that receives read/write requests from an application or a file system, and relays those requests to a storage device. BDL-APT automatically optimizes the number of IP-SAN sessions based on the measured network status using our parallelism tuning mechanism based on a numerical computation algorithm, Golden Section Search method. We perform preliminarily investigation on the effectiveness of BDL-APT in realistic network environments using our BDL-APT implementation. Consequently, we demonstrate that our BDL-APT operates effectively in long-fat networks.

No abstract

Many TOE (TCP/IP Offload Engine) devices have been developed and installed in high-end systems.  Although there have been many analysis of the effectiveness of TOE devices, there remain open issues related to them.  For example, it has not been clarified which part of TCP/IP processing should be performed with hardware and which with software, or how end-to-end TCP/IP performance is affected by the introduction of a TOE device.  In this paper, we propose VOSE (Virtual Offloading with Software Emulation), which is a technique for measuring TCP/IP performance improvements derived from different type of TOE devices without implementing TOE prototypes really.  VOSE enables virtual offloading without requiring a hardware TOE device by virtually emulating TOE processing on both source and destination end hosts.  For demonstrating the effectiveness of VOSE, we apply VOSE to the TCP checksum and IPsec protocol.  We extensively examine the accuracy of virtual offloading with VOSE, by comparing performance (i.e., end-to-end performance and CPU processing overhead) between VOSE and a dedicated TOE device.  Moreover, we estimate performance improvement that are derived from several TOE devices of IPsec and combinations of those devices, by applying VOSE to header authenticating and payload encryption in IPsec protocol.  Consequently, we show that performance improvements which are derived from TOE devices can be estimated correctly.

In this paper we propose Block Device Layer with Automatic Parallelism Tuning (BDL-APT), a mechanism that maximizes the goodput of heterogeneous IP-based Storage Area Network (IP-SAN) protocols in long-fat networks. BDL-APT parallelizes data transfer using multiple IP-SAN sessions at a block device layer on an IP-SAN client, automatically optimizing the number of active IP-SAN sessions according to network status. A block device layer is a layer that receives read/write requests from an application or a file system, and relays those requests to a storage device. BDL-APT parallelizes data transfer by dividing aggregated read/write requests into multiple chunks, then transferring a chunk of requests on every IP-SAN session in parallel. BDL-APT automatically optimizes the number of active IP-SAN sessions based on the monitored network status using our parallelism tuning mechanism. We evaluate the performance of BDL-APT with heterogeneous IP-SAN protocols (NBD, GNBD and iSCSI) in a long-fat network. We implement BDL-APT as a layer of the Multiple Device driver, one of major software RAID implementations included in the Linux kernel. Through experiments, we demonstrate the effectiveness of BDL-APT with heterogeneous IP-SAN protocols in long-fat networks regardless of protocol specifics.