Saeyoung Han

A Performance Evaluation Methodology in Virtual Environments

System virtualization come into the spotlight again with the emerging virtual machine (VM) technologies (e.g., Parallels, VMware and Xen). With these market trends toward virtual environments, many research groups are developing evaluation tools to check the performance of virtual systems and their overheads. However, the performance characterization in virtual environments has not been established yet for many challenging issues. In this paper, we categorize four virtualization models for characterizing the performance of virtual environments and present an efficient evolution methodology to compare the performance of virtualized versus non-virtualized systems.

An NAT-Based Communication Relay Scheme for Private-IP-Enabled MPI over Grid Environments

In this paper we propose a communication relay scheme combining the NAT and a user-level proxy to support private IP clusters in Grid environments. Compared with the user-level two-proxy scheme used in PACX-MPI and Firewall-enabled MPICH-G, the proposed scheme shows performance improvement in terms of latency and bandwidth between the nodes located in two private IP clusters. Since the proposed scheme is portable and provides high performance, it can be easily applied to any private IP enabled solutions including the private IP enabled MPICH solution for Globus toolkit.

A CPU provision scheme considering virtual machine scheduling delays in Xen virtualized environment

Managing hardware resources efficiently when they are shared in virtualized environment is a critical issue for improving application QoS. For CPU resources, virtual machine scheduling delays should be additionally concerned, because they cause unexpected service delays to guest applications and can consequently violate the application QoS. There have been many researches to manage CPU resources by monitoring resource utilization and analyzing application performance via mathematical techniques such as queuing models. However, most of the studies cannot always guarantee the application QoS, because they do not reflect the virtual machine scheduling delay. In this paper, we propose a CPU provision scheme considering virtual machine scheduling delays in Xen environment. We compare the performance of the proposed scheme with that of Sandpiper and show that the proposed scheme decreases the number of SLA violations.

A QoS based migration scheme for virtual machines in data center environments

We propose a QoS based migration system for virtual machines in data center environments. In order for our QoS based migration scheme to ensure QoS requirements for virtual machines, it estimates the QoS during migrations for each virtual machine, and selects the virtual machine that is most likely not to violate its QoS requirement. Moreover, it improves the overall system utilization by choosing the destination server for migration to balance resource usages.

A Dynamic Layer Management Scheme for a Superpeer Ring with a Loosely-Consistent DHT

We propose a dynamic layer management scheme for superpeer ring with a loosely-consistent DHT, where high capacity peers are selected as superpeers based on the capacity data colleted by mobile agents. The benchmarking results show that our scheme outperforms the static superpeer systems in terms of search latency at the expense of overhead.

An efficient search scheme using self-organizing hierarchical ring in unstructured peer-to-peer systems

We propose an efficient search scheme based on a self-organizing hierarchical ring structure in unstructured peer-to-peer systems Our solution maintains a consistent super-peer ratio and makes the peers with relatively high capacities super-peers in dynamic environments The benchmarking results show that the proposed algorithm outperforms the static algorithm that has the fixed number of rendezvous peers such as JXTA.

A content-based load balancing algorithm for metadata servers in cluster file systems

A metadata service is one of the important factors to affect the performance of cluster file systems. We propose a content-based load balancing algorithm that dynamically distributes client requests to appropriate metadata servers based on the types of metadata operations. By replicating metadata and logging update messages in each server rather than moving metadata across servers, we significantly reduce the response time and evenly distribute client requests among metadata servers.

A dynamic layer management scheme for a superpeer ring with a loosely-consistent DHT

In this paper we propose a Dynamic Layer Management (DLM) scheme for a self-organising superpeer ring with a loosely-consistent DHT. Each peer estimates the global capacity based on the information collected by mobile agents and autonomously decides whether it can be a superpeer or not. The proposed scheme also changes the size of the superpeer ring adaptively according to the size of the network, and maintains an appropriate number of superpeers in dynamic peer-to-peer environments. The benchmarking results show that our scheme outperforms the fixed superpeer system in terms of search latency at the expense of message overhead added for the DLM.

Adapting Superpeer Size Using Particle Swam Optimization for Self-Organizing Superpeer Ring with Loosely Consistent DHT

A superpeer size adaptation using PSO is proposed for the self-organizing superpeer ring with loosely-consistent DHT. In the proposed scheme, all peers memorize the superpeer ratio where they or their neighbors experienced the best response time and adjust the superpeer ratio autonomously using PSO algorithm. Without tuning the optimal superpeer ratio value manually, the self-organizing superpeer ring can maintain moderately good response time in dynamic peer-to-peer systems.

A Peer-to-Peer Search Scheme using Self-Organizing Super Peer Ring

We propose a peer-to-peer search scheme in which super peers with high rapacity constitute a ring by themselves and all peers utilize the ring for their query and publishing keys. In a dynamic peer-to-peer environment, the size of the super peer ring changes adaptively according to the changes of the system, but more powerful peers are continuously promoted to super peers. By simulations, we show that our proposed search scheme improves the query success rate of Gnutella+1hop replication search method, and maintains shorter query delay than JXTA as a static ring.