Jørgen Sværke Hansen

Optimizing TCP forwarder performance

A TCP forwarder is a network node that establishes and forwards data between a pair of TCP connections. An example of a TCP forwarder is a firewall that places a proxy between a TCP connection to an external host and a TCP connection to an internal host, controlling access to a resource on the internal host. Once the proxy approves the access, it simply forwards data from one connection to the other. We use the term TCP forwarding to describe indirect TCP communication via a proxy in general. This paper characterizes the behavior of TCP forwarding, and illustrates the role TCP forwarding plays in common network services like firewalls and HTTP proxies. We then introduce an optimization technique, called connection splicing, that can be applied to a TCP forwarder, and report the results of a performance study designed to evaluate its impact. Connection splicing improves TCP forwarding performance by a factor of two to four, making it competitive with IP router performance on the same hardware.

Using idle disks in a cluster as a high-performance storage system

In many clusters today, the local disks of a node are only used sporadically. This paper describes the software support for sharing of disks in clusters, where the disks are distributed across the nodes in the cluster, thereby allowing them to be combined into a high-performance storage system. Compared to centralized storage servers, such an architecture allows the total I/O capacity of the cluster to scale up with the number of nodes and disks. Additionally, our software allows customizing the functionality of the remote disk access using a library of code modules. A prototype has been implemented on a cluster connected by a Scalable Coherent Interface (SCI) network and performance measurements using both raw device access and a distributed file system show that the performance is comparable to dedicated storage systems and that the overhead of the framework is moderate even during high load. Thus, the prospects are that clusters sharing disks distributed among the nodes will allow both the application processing power and total I/O capacity of the cluster to scale up with the number of nodes.

A distributed shared buffer space for data-intensive applications

Efficient memory allocation and data transfer for cluster-based data-intensive applications is a difficult task. Both changes in cluster interconnects and application workloads usually require timing of the application and network code. We propose separating control and data transfer traffic by accessing data through a DSM-like cluster-wide shared buffer space and only including buffer references in the control messages. Using a generic API for accessing buffers allows for tuning data transfer without changing the application code. A prototype, implemented in the context of a distributed storage system, has been validated with several networking technologies, showing that such a framework can combine performance and flexibility.

Simplifying administration through dynamic reconfiguration. in a cooperative cluster storage system

Cluster storage systems where storage devices are distributed across a large number of nodes are able to reduce the I/O bottleneck problems present in most centralized storage systems. However, such distributed storage devices are hard to manage efficiently. In this paper, we examine the use of explicit, component-based (command and data) paths between hosts and disks as a vehicle for performing nondisruptive storage system reconfiguration. We describe the mechanisms necessary to perform reconfigurations and show how they can be used to handle two management tasks: migration between network technologies and rebuilding a disk in a mirror. Our approach is validated through initial performance measurements of these two tasks using a prototype implementation. The results show that online reconfiguration is possible at a modest cost

Prioritizing Network Event Handling in Clusters of Workstations

The use of modern system area networking technologies [9,3] to construct tightly integrated clusters of workstations exposes two weaknesses of current operating systems. First, the low latency of current networks is often hidden from the application due to the high cost of interrupt handling. Second, network event handling during high load may result in serious performance degradation because all processor time is used for network event handling resulting in application starvation. This paper concerns the problems related to providing efficient and stable network event handling for clusters of workstations and network servers. By stable we mean that the throughput and response time of the system does not suffer when the workload offered to the system is increased beyond the maximum capacity of the system.

Implementing a File System Interface to SCI

This chapter deals with the issues in implementing a file system interface to the shared memory in an SCI cluster. We describe the possibilities of sharing in file systems and how it can be implemented for SCI in UNIX systems and Windows NT. We present our prototype, SciOS, which implements a memory-based distributed file system. We find that the file system interface integrates SCI well with the operating system and provides sharing mechanisms, a symbolic name space, and possibilities for protection on multi-user SCI clusters.