Chang Soo Kim

Volume management in SAN environment

Logical volume managers have long been key components of a storage system. Their key features are creation of logical or virtual views of physical storage devices and support for various software RAID levels. These make it possible to overcome the limits to capacity, availability and performance of a physical storage device. Most logical volume managers are operated in a single system environment. They are not adequate for SAN (storage area network) environments where several hosts share and access a logical volume at the same time. Some recent logical volume managers are run in a multi-host environment. However, they cannot support the enterprise computing environments in which the system must support 24/spl times/7/spl times/365 uptime operations such as online resizing and online backup. We propose a logical volume manager called SANtopia Volume Manager that supports multihost environments and provides various volume management features to support enterprise computing. Also it is a cluster enabled logical volume manager that maximizes the parallelism for high performance, and provides high scalability and high availability.

Lightweight live migration for high availability cluster service

High availability is a critical feature for service clusters and cloud computing, and is often considered more valuable than performance. One commonly used technique to enhance the availability is live migration, which replicates services based on virtualization technology. However, continuous live migration with checkpointing will introduce significant overhead. In this paper, we present a lightweight live migration (LLM) mechanism to integrate wholesystem migration and input replay efforts, which aims at reducing the overhead while providing comparable availability. LLM migrates service requests from network clients at high frequency during the interval of checkpointing system updates. Once a failure happens to the primary machine, the backup machine will continue the service based on the virtual machine image and network inputs at their respective last migration rounds. We implemented LLM based on Xen and compared it with Remus--a state-of-the-art effort that enhances the availability by checkpointing system status updates. Our experimental evaluations show that LLM clearly outperforms Remus in terms of network delay and overhead. For certain types of applications, LLM may also be a better alternative in terms of downtime than Remus. In addition, LLM achieves transaction level consistency like Remus.

VPC: Scalable, Low Downtime Checkpointing for Virtual Clusters

A virtual cluster (VC) consists of multiple virtual machines (VMs) running on different physical hosts, inter-connected by a virtual network. A fault-tolerant protocol and mechanism are essential to the VCs availability and usability. We present Virtual Predict Check pointing (or VPC), a lightweight, globally consistent check pointing mechanism, which checkpoints the VC for immediate restoration after VM failures. By predicting the checkpoint-caused page faults during each check pointing interval, VPC further reduces the solo VM downtime than traditional incremental check pointing approaches. Besides, VPC uses a globally consistent check-pointing algorithm, which preserves the global consistency of the VMs execution and communication states, and only saves the updated memory pages during each check pointing interval to reduce the entire VC downtime. Our implementation reveals that, compared with past VC check pointing/migration solutions including VNsnap, VPC reduces the solo VM downtime by as much as 45%, under the NPB benchmark, and reduces the entire VC downtime by as much as 50%, under the NPB distributed program. Additionally, VPC incurs a memory overhead of no more than 9%. In all cases, VPCs performance overhead is less than 16%.

Enhancing the performance of high availability lightweight live migration

Remus is one of the first systems which implemented whole virtual machine replication to achieve high availability (HA). Recently a fast, lightweight migration mechanism (LLM) was proposed to reduce the long network delay in Remus. However, these virtualized systems have the long downtime problem, which is a bottleneck to achieve HA. Based on LLM, in this paper, we describe a fine-grained block identification (or FGBI) mechanism to reduce the downtime in virtualized systems so as to achieve HA, with support for a block sharing mechanism and hybrid compression method. We implement the FGBI mechanism and evaluate it against LLM and Remus, using several benchmarks such as Apache, SPECweb, NPB and SPECsys. Our experimental results reveal that FGBI reduces the type I downtime over LLM and Remus by as much as 77% and 45% respectively, and reduces the type II downtime by more than 90% and more than 70%, compared with LLM and Remus respectively. Moreover, in all cases, the performance overhead of FGBI is less than 13%.

An Efficient Global Request Dispatching Scheme for Multimedia Streaming Services

For providing high quality multimedia streaming services over high speed network, various solutions such as performance improvement of computer system, efficient cluster management and request dispatching into back-end nodes, good load balancing scheme, global infrastructure management and efficient contents distribution mechanism are required. In this paper, we will describe the global request dispatching scheme in design and implementation aspects for supporting high quality multimedia streaming services in high speed network environments. With this scheme, we can get good performance and correctness in dispatching service requests.

Design of system for multimedia streaming service

In the current life, Internet is playing a key role in transferring diverse information. As the performance of Internet has been growing, transferred information has the form of multimedia streams such as video and audio. However, in current computing and internet environment, multimedia streaming services can’t be accomplished smoothly due to the various factors such as internet bandwidth, computing power and so on. In this paper, we are proposing an efficient system including hardware and software in terms of the design concept and system architecture for the multimedia streaming service. We also present the implementation result for the system.

A JIT Compilation-Based Unified SQL Query Optimization System

In-memory databases are gaining attention as a solution to efficiently support SQL queries on large volume of data, as main memories are becoming cheaper and grow in size. However, their query performance is not well improved on modern hardware with faster CPUs, registers and caches due to the limitation of the classical iterator style query processing model. We propose a unified SQL query optimization system using JIT compilation of OLTP, OLAP, and Stored Procedure workloads for enhanced performance on modern hardware.

On Multiple Description Streaming with Cluster-based Server

The cluster system with shared-nothing storage cannot escape from the problem of skewed request toward specific contents. This paper, therefore, suggests streaming method using MDC (multiple description coding) instead of using single original content; this method is able to cope with skewed request in shared-nothing storage server as well as to continue to provide services in case of the system failure. Also, the system can support adaptive streaming service according to user player type, network status, the load of server, and client

The SZIT based-mapping method for on-line adding disks in software RAID

According as the users can connect to remote hosts at anytime by using internet, the demand of non-stop system is increased. Also, the research about on-line adding/deleting disks and hosts are progressed. This paper suggests that the method of reducing additional operation when we add disks to the stripping system. That method can fast mapping like equation based mapping method because of using so-called SZIT(Striping Zone Information Table) and equations. And that method can reduce the relocation overhead in striping system at adding disks.