Sung-In Jung

Inter-domain socket communications supporting high performance and full binary compatibility on Xen

Communication performance between two processes in their own domains on the same physical machine gets improved but it does not reach our expectation. This paper presents the design and implementation of high-performance inter-domain communication mechanism, called XWAY, that maintains binary compatibility for applications written in standard socket interface. As a result of our survey, we found that three overheads mainly contribute to the poor performance; those are TCP/IP processing cost in each domain, page flipping overhead, and long communication path between both sides of a socket. XWAY achieves high performance by bypassing TCP/IP stacks, avoiding page flipping overhead, and providing a direct, accelerated communication path between domains in the same machine. Moreover, we introduce the XWAY socket architecture to support full binary compatibility with as little effort as possible. We implemented our design on Xen 3.0.3-0 with Linux kernel 2.6.16.29, and evaluated basic performance, the speed of file transfer, DBT-1 benchmark, and binary compatibility using binary image of real socket applications. In our tests, we have proved that XWAY realizes the high performance that is comparable to UNIX domain socket and ensures full binary compatibility. The basic performance of XWAY, measured with netperf, shows minimum latency of 15.6 usec and peak bandwidth of 4.7Gbps, which is superior to that of native TCP socket. We have also examined whether several popular applications using TCP socket can be executed on XWAY with their own binary images. Those applications worked perfectly well.

SOVIA: a user-level sockets layer over virtual interface architecture

The Virtual Interface Architecture (VIA) is an industry standard user-level communication architecture for system area networks. The VIA provides a protected, directly-accessible interface to a network hardware, removing the operating system from the critical communication path. In this paper, we design and implement a user-level Sockets layer over VIA, named SOVIA (Sockets Over VIA). Our objective is to use the SOVIA layer to accelerate the existing Sockets-based applications with a reasonable effort and to provide a portable and high performance communication library based on VIA to the application developers.SOVIA realizes comparable performance to native VIA, showing the minimum latency of 10.5usec and the peak bandwidth of 814Mbps on Giganets cLAN. We have verified the functional compatibility with the existing Sockets API by porting FTP (File Transfer Protocol) and RPC (Remote Procedure Call) applications over the SOVIA layer. Compared to the Giganets LANE driver which emulates TCP/IP inside the kernel, SOVIA easily doubles the file transfer bandwidth in FTP and reduces the latency of calling an empty remote procedure by 77% in RPC applications.

GNBD/VIA: a network block device over virtual interface architecture on Linux

This paper describes a design and implementation of GNBD/VIA, a network block device (NBD) over virtual interface architecture (VIA), and evaluates its performance on Linux-based cluster of PCs. VIA is a user-level memory-mapped communication model which provides zero-copy communication by removing the operating system from the critical communication path. Typically, an NBD layer offers the abstraction of a storage media across the network. GNBD/VIA attempts to improve the performance of the NBD layer by employing the lightweight VIA communication mechanisms between NBD servers and clients. To our best knowledge, GNBD/VIA is the first implementation of NBD on VIA. GNBD/VIA outperforms the normal NBD placed on top of TCP/IP protocol stacks, and achieves the performance comparable to local disk devices, showing the read (write) bandwidth of 30.6MB/s (25.9MB/s) on the evaluation platform with UDM.9100 hard disks and Emulex cLAN adapters.

Building a high-performance communication layer over virtual interface architecture on Linux clusters

The Virtual Interface Architecture (VIA) is an industry standard user-level communication architecture for cluster or system area networks. The VIA provides a protected, directly-accessible interface to a network hardware, removing the operating system from the critical communication path. Although the VIA enables low-latency high-bandwidth communication, the application programming interface defined in the VIA specification lacks many high-level features. In this paper, we develop a high performance communication layer over VIA, named SOVIA (Sockets Over VIA). Our goal is to make the SOVIA layer as efficient as possible so that the performance of native VIA can be delivered to the application, while retaining the portable Sockets semantics. We find that the single-threaded implementation with conditional sender-side buffering is effective in reducing latency. To increase bandwidth, we implement a flow control mechanism similar to the TCPs sliding window protocol. Our flow control mechanism is enhanced further by adding delayed acknowledgments and the ability to combine small messages. With these optimizations, SOVIA realizes comparable performance to native VIA, showing the latency of 10.5 µsec for 4-byte messages and the peak bandwidth of 814Mbps on Giganets cLAN. The functional validity of SOVIA is verified by porting FTP (File Transfer Protocol) application over SOVIA.

A Log-based Flash Translation Layer for Large NAND flash memory

Flash memory has been widely used in mobile and embedded systems because of non-volatility, low power consumption, shock resistance, lightweight, small size, and fast access speed. However, flash memory has two disadvantages. One is that data cannot be overwritten before erase in advance. The other is that each erasable unit of flash memory has the limitation of possible number of erase operations: generally 100,000 program/erase cycles. To address these problems, the systems have a translation layer and a garbage collection mechanism to provide applications a transparent storage service. It called a flash translation layer (FTL). Recently, mobile and embedded systems use many types of multimedia data. It means that the storage system must have a long life time, fast access speed, and large capacity. To satisfy these requirements, we propose new FTL algorithm called LogFTL (Log-based FTL). We provide the performance result based on our implementation of LogFTL and other FTLs

Implementation and evaluation of EXT3NS multimedia file system

The EXT3NS is a scalable file system designed to handle video streaming workload in large-scale on-demand streaming services. It is based on a special H/W device, called Network-Storage card (NS card), which aims at accelerating streaming operation by shortening the data path from storage device to network interface. The design objective of EXT3NS is to minimize the delay and the delay variance of I/O request in the sequential workload on NS card. Metadata structure, file organization, metadata structure, unit of storage, etc. are elaborately tailored to achieve this objective. Further, EXT3NS provides the standard APIs to read and write files in storage unit of NS card. The streaming server utilizes it to gain high disk I/O bandwidth, to avoid unnecessary memory copies on the data path from disk to network, and to alleviates CPUs burden by offloading parts of network protocol processing, The EXT3NS is a full functional file system based on the popular EXT3. The performance measurements on our prototype video server show obvious performance improvements. Specifically, we obtain better results from file system benchmark program, and obtain performance improvements in disk read and network transmission, which leads to overall streaming performance increase. Especially, the streaming server shows much less servers CPU utilization and less fluctuation of client bit rate, hence more reliable streaming service is possible.

Design and implementation of a user-level Sockets layer over Virtual Interface Architecture

The Virtual Interface Architecture (VIA) is an industry standard user‐level communication architecture for system area networks. The VIA provides a protected, directly‐accessible interface to a network hardware, removing the operating system from the critical communication path. In this paper, we design and implement a user‐level Sockets layer over VIA, named SOVIA (Sockets Over VIA). Our objective is to use the SOVIA layer to accelerate the existing Sockets‐based applications with a reasonable effort and to provide a portable and high‐performance communication library based on VIA to application developers.

Cloud service broker portal: Main entry point for multi-cloud service providers and consumers

Cloud-service-broker needs a virtual service portal between multiple cloud-service-providers and cloud-service-consumers. The cloud-service-broker portal enables the cloud-service-providers to specify available their services. In addition, the cloud-service-consumers may find the most suitable services by negotiating the agreements on the services. The cloud-service-broker as an emerging technology intermediates heterogeneous multiple cloud services for both the providers and consumers. In this paper, we suggest the web-based user interface design of the cloud-service-broker portal to support different providers and consumers.

The new approach for inter-communication between guest domains on Virtual Machine Monitor

In general, the speed of internal communication on a single-machine is faster than the other communication.(e.g. communication between remote PCs through the local network) Therefore, we expected that the internal communication between guest domains on a single-machine of virtual machine monitor (VMM)[1] was also fast. However, in the experiment, it showed that rather the speed was slower than the other communication because each guest domain was not well considered to be executed on VMM. In other words, the guest domain has some problems in the architecture of communication and data transmission on VMM. In this paper, we find the causes of those problems and suggest a new design for internal communication to improve its performance. It is well considered to be executed on VMM and shows the performance improvement of bandwidth up to 8 times in comparison with existing internal communication. And it also reduced the latency to 25%.

Design and implementation of zero-copy data path for efficient file transmission

Huge requests for file transmission by concurrent users cause excessive memory copy operations and system calls between kernel and user space, which are very expensive and require many CPU cycles for processing. And it has been a bottleneck that limits the number of serviceable requests and prevents CPU resources from being allocated to other processes. In this paper, we suppose the zero-copy data path for efficient file transmission to solve the upper described problems. To do that, we used existing system call interface, sendfile, for file transmission from one device to another and used the same buffer page to transmit a file. The supposed zero-copy data path reduces CPU cost per transaction by eliminating CPU copies between kernel and user space and by avoiding unnecessary system calls in user space. The improved CPU efficiency means that a larger number of requests can be serviced with the same CPU configuration.