A dynamic approach for efficient TCP buffer allocation

2002

TCP Server is a system architecture aiming to of- fload network processing from the host(s) running an Internet server. The basic idea is to execute the TCP/IP processing on a dedicated processor, node, or device (the TCP server) using low-overhead, non-intrusive communication between it and the host(s) running the server application. In this paper, we propose, implement, and eval- uate

Proceedings Third IEEE International Symposium on Object-Oriented Real-Time Distributed Computing (ISORC 2000) (Cat. No. PR00607), 2000

There have been many debates about the feasibility of providing guaranteed Quality of Service (QoS) when network traffic travels beyond the enterprise domain and into the vast unknown of the Internet. Many mechanisms have been proposed to bring QoS to TCP/IP and the Internet (RSVP, DiffServ, 802.1p). However, until these techniques and the equipment to support them become ubiquitous, most enterprises will rely on local prioritization of the traffic to obtain the best performance for mission critical and time sensitive applications. This work explores prioritizing critical TCP/IP traffic using a multi-queue buffer management strategy that becomes biased against random low priority flows and remains biased while congestion exists in the network. This biasing implies a degree of unfairness but proves to be more advantageous to the overall throughput of the network than strategies that attempt to be fair. Only two classes of services are considered where TCP connections are assigned to these classes and mapped to two underlying queues with round robin scheduling and shared memory. In addition to improving the throughput, cell losses are minimized for the class of service (queue) with the higher priority.

2001

10th IEEE Symposium on Computers and Communications (ISCC'05)

A packet buffer for a protocol processor is a large shared memory space that holds incoming data packets in a computer network. This paper investigates four packet buffer management algorithms for a protocol processor including Dynamic Algorithm with Different Thresholds (DADT), which is proposed to reduce the packet loss ratio efficiently. The proposed algorithm takes the advantage of different packet sizes for each application by allocating buffer space for each queue proportionally. According to our simulation results, the DADT algorithm works well in reducing packet loss ratio compared to other three algorithms.

2001

With the widespread arrival of bandwidth-intensive applications such as bulk-data transfer, multi-media web streaming and computational grids for high-performance computing, networking performance over the wide-area network has become a critical component in the infrastructure. Tragically, operating systems are still tuned for yesterday's WAN speeds and network applications. As a result, a painstaking process of manually tuning system buffers must be undertaken to make TCP flow-control scale to meet the needs of today's bandwidth-rich networks. Consequently, we propose an operating system technique called dynamic right-sizing that eliminates the need for this manual process. Previous work has also attacked this problem, but with less than complete solutions. Our solution is more efficient, more transparent, and applies to a wider set of applications, including those that require strict flow-control semantics because of performance disparities between the sender and receiver.

1998

The rapid growth of the World Wide Web in recent years has caused a significant shift in the composition of Internet traffic. Although past work has studied the behavior of TCP dynamics in the context of bulk-transfer applications and some studies have begun to investigate the interactions of TCP and HTTP, few have used extensive realworld traffic traces to examine the problem. This interaction is interesting because of the way in which current Web browsers use TCP connections: multiple concurrent short connections from a single host.

Journal of Computer Networks and Communications

In the absence of losses, TCP constantly increases the amount of data sent per instant of time. This behavior leads to problems that affect its performance, especially when multiple devices share the same gateway. Several studies have been done to mitigate such problems, but many of them require TCP side changes or a meticulous configuration. Some studies have shown promise, such as the use of gateway techniques to change the receiver’s advertised window of ACK segments based on the amount of memory in the gateway; in this work, we use the term “network-return” to refer to these techniques. In this paper, we present a new network-return technique called early window tailoring (EWT). For its use, it does not require any modification in the TCP implementations at the sides and does not require that all routers in the path use the same congestion control mechanism, and the use in the gateway is sufficient. With the use of the simulator ns-3 and following the recommendations of RFC 7928...

Journal of Grid Computing, 2003

It is often claimed that TCP is not a suitable transport protocol for data intensive Grid applications in high-performance networks. We argue that this is not necessarily the case. Without changing the TCP protocol, congestion control, or implementation, we show that an appropriately tuned TCP bulk transfer can saturate the available bandwidth of a network path. The proposed technique, called SOBAS, is based on automatic socket buffer sizing at the application layer. In non-congested paths, SOBAS limits the socket buffer size based on direct measurements of the received throughput and of the corresponding round-trip time. The key idea is that the send window should be limited, after the transfer has saturated the available bandwidth in the path, so that the transfer does not cause buffer overflows (‘self-induced losses’). A difference with other socket buffer sizing schemes is that SOBAS does not require prior knowledge of the path characteristics, and it can be performed while the transfer is in progress. Experimental results in several high bandwidth-delay product paths show that SOBAS provides consistently a significant throughput increase (20% to 80%) compared to TCP transfers that use the maximum possible socket buffer size. We expect that SOBAS will be mostly useful for applications such as GridFTP in non-congested wide-area networks.

Computer Science Department, Rutgers University, 2002

TCP Server is a system architecture aiming to offload network processing from the host (s) running an Internet server. The basic idea is to execute the TCP/IP processing on a dedicated processor, node, or device (the TCP server) using low-overhead, non-intrusive communication between it and the host (s) running the server application. In this paper, we propose, implement, and evaluate three TCP Server architectures:(1) a dedicated network processor on a symmetric multiprocessor (SMP) server,(2) a dedicated node on a cluster- ...

2002

Rather than painful, manual, static, per-connection optimization of TCP buffer sizes simply to achieve acceptable performance for distributed applications [8, 10], many researchers have proposed techniques to perform this tuning automatically [4, 7, 9, 11, 12, 14]. This paper first discusses the relative merits of the various approaches in theory, and then provides substantial experimental data concerning two competing implementations-the buffer autotuning already present in Linux 2.4.x and "Dynamic Right-Sizing." This paper reveals heretofore unknown aspects of the problem and current solutions, provides insight into the proper approach for different circumstances, and points toward ways to further improve performance.

2001

This paper proposes a method for improving the performance of web servers servicing static HTTP requests. The idea is to give preference to those requests which are short, or have small remaining processing requirements, in accordance with the SRPT (Shortest Remaining Processing Time) scheduling policy.

Proceedings of the 2012 ACM workshop on Capacity sharing - CSWS '12, 2012

Today, content replication methods are common ways of reducing the network and servers load. Present content replication solutions have different problems, including the need for pre-planning and management, and they are ineffective in case of sudden traffic spikes. In spite of these problems, content replication methods are more popular today than ever, simply because of an increasing need for load reduction. In this paper, we propose a shared buffering model that, unlike current proxy-based content replication methods, is native to the network and can be used to alleviate the stress of sudden traffic spikes on servers and the network. We outline the characteristics of a new transport protocol that uses the shared buffers to offload the server work to the network or reduce the pressure on the overloaded links.

2002

The growth of the Internet and the deployment of new Web-based services have placed a high emphasis on system performance. Providers are interested in low-cost architectures and techniques to reduce the response times perceived by users during HTTP transactions. A widespread approach is based on the use of distributed Web server architectures based on clusters of servers driven by dedicated machines that take charge of dispatching the HTTP requests. An increasingly popular mechanism to carry out dispatching is based on the analysis of request content. This is typically realized at the application level, since an implementation at the transport level may result difficult due to the connection oriented nature of TCP. This paper proposes a modification to TCP in order to support content-aware scheduling in the kernel of an OS, and analyzes the advantages of such a scheduling approach by implementing a name-based algorithm and comparing its performance with the ones delivered by content-blind scheduling algorithms.

2000

We discuss the performance effects of using per-transaction TCP connections for HTTP access, and the proposed optimizations of avoiding per-transaction re-connection and TCP slow-start restart overheads. We analyzed the performance penalties of the interaction of HTTP and TCP. Our observations indicate that the proposed optimizations do not affect Web access for the vast majority of users. Most users see end-to-end

2012

The bufferbloat is a serious problem of the modern nets with complex organization. The persistent saturation of the net with an excessive buffering on hops results in a huge latency which can make interactive and realtime connections practically unusable. Most research in this area is devoted to Active Queue Management (AQM) which is used to keep the packet queue size correct on the bottleneck of the network path. This article gives the other look on the bufferbloat. It suggests to fight with the congestion-the reason of buffer saturation. It states that currently most widely used TCP loss-based congestion control algorithms have the problem which is shown in impossibility to correctly estimate the bandwidth in the presence of excessive buffering, and suggests to use the delay-based approach for congestion avoidance. It presents ARTCP which is delay-based and does not use burst sends limited by the congestion window, but transmits data on the adaptive rate estimated with pacing intervals.