One Server Per City: Using TCP for Very Large SIP Servers

Ono, Kumiko; Schulzrinne, Henning

doi:10.1007/978-3-540-89054-6_7

Cited by 13 publications

(9 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They also conclude that authentication considerably affects the performance of the proxy. Ono et al [18] analyze how the use of TCP affects the performance and scalability of a proxy proposing several configuration changes to improve performance. Cortes et al [8] evaluate the parsing, string processing, memory allocation, thread overhead and overall capacity of a proxy.…”

Section: Related Workmentioning

confidence: 99%

Improving authentication performance of distributed SIP proxies

Dacosta

Balasubramaniyan

Ahamad

et al. 2009

Proceedings of the 3rd International Conference on Principles, Systems and Applications of IP Telecommunications

View full text Add to dashboard Cite

The performance of SIP proxies is critical for the robust operation of many applications. However, the use of even light-weight authentication schemes can significantly degrade throughput in these systems. In particular, systems in which multiple proxies share a remote authentication database can experience reduced performance due to latency. In this paper, we investigate how the application of parallel execution and batching can be used to maximize throughput while carefully balancing demands for bandwidth and call failure rates. Through the use of a modified version of OpenSER, a high-performance SIP proxy, we demonstrate that the traditional recommendation of simply launching a large number of parallel processes not only incurs substantial overhead and increases dropped calls, but can actually decrease call throughput. An alternative technique that we implement, request batching, fails to achieve similarly high proxy throughput. Through a carefully selected mix of batching and parallelization, we reduce the bandwidth required to maximize authenticated signaling throughput by the proxy by more than 75%. This mix also keeps the call loss rates below 1% at peak performance. Through this, we significantly reduce the cost and increase the throughput of authentication for large-scale networks supporting SIP applications.

show abstract

Section: Related Workmentioning

confidence: 99%

Improving authentication performance of distributed SIP proxies

Dacosta

Balasubramaniyan

Ahamad

et al. 2009

Proceedings of the 3rd International Conference on Principles, Systems and Applications of IP Telecommunications

View full text Add to dashboard Cite

show abstract

“…Within the rate-based methods, downstream server has a consistent supervision over the rate of messages delivered from total upstream servers and informs its admissible rate to them [5,20]. In loss-based methods, downstream server repeatedly measures its current load and, accordingly, requests the upstream servers to reduce their transmitted load [21]. In window-based mechanisms, the load is not transmitted to the downstream server unless there are empty slots in upstream server window.…”

Section: Distributedmentioning

confidence: 99%

Overload mitigation mechanism for VoIP networks: a transport layer approach based on resource management

Montazerolghaem

Seno

Yaghmaee

et al. 2016

Trans. Emerging Tel. Tech.

View full text Add to dashboard Cite

Session initiation protocol (SIP) is an application layer signalling protocol for the set‐up, management and termination of multimedia networks such as voice over IP standardised by Internet Engineering Task Force. This protocol has also been recognised by International Telecommunication Union as a main core of next‐generation networks. When this network is under overload for reasons such as improper design, instantaneous swarm, components error and sudden reduction of processing capacity, its efficiency is considerably reduced. Overload occurs when the SIP proxy lacks sufficient CPU and memory resources to process all messages. Because the overload cannot be prevented completely, it is important to equip SIP proxies with an effective overload mitigation mechanism. In this paper, capabilities of transmission control protocol are used in transport layer to reduce SIP proxy overload through the proper allocation of proxy resources. Major activities in this field assume SIP on user datagram protocol, which practically does not result in optimal throughput. To evaluate this approach, an Asterisk open source proxy is used. Our implementation results in the real testbed indicate the efficiency improvement of Asterisk proxy under overload. Copyright © 2016 John Wiley & Sons, Ltd.

show abstract

“…At 10 million concurrent connections, that amounts to 580 megabytes. In contrast, the Linux kernel uses in excess of 2 kilobytes per TCP connection [8]-not counting buffers for storing packet fragments. masscan is two orders of magnitude more efficient in its use of memory!…”

Section: B Memory Overheadmentioning

confidence: 99%

Finite State Machine Parsing for Internet Protocols: Faster Than You Think

Graham¹,

Johnson

2014

2014 IEEE Security and Privacy Workshops

View full text Add to dashboard Cite

Abstract-A parser's job is to take unstructured, opaque data and convert it to a structured, semantically meaningful format. As such, parsers often operate at the border between untrusted data sources (e.g., the Internet) and the soft, chewy center of computer systems, where performance and security are paramount. A firewall, for instance, is precisely a trust-creating parser for Internet protocols, permitting valid packets to pass through and dropping or actively rejecting malformed packets. Despite the prevalence of finite state machines (FSMs) in both protocol specifications and protocol implementations, they have gained little traction in parser code for such protocols. Typical reasons for avoiding the FSM computation model claim poor performance, poor scalability, poor expressibility, and difficult or time-consuming programming.In this research report, we present our motivations for and designs of finite state machines to parse a variety of existing Internet protocols, both binary and ASCII. Our hand-written parsers explicitly optimize around L1 cache hit latency, branch misprediction penalty, and program-wide memory overhead to achieve aggressive performance and scalability targets. Our work demonstrates that such parsers are, contrary to popular belief, sufficiently expressive for meaningful protocols, sufficiently performant for high-throughput applications, and sufficiently simple to construct and maintain. We hope that, in light of other research demonstrating the security benefits of such parsers over more complex, Turing-complete codes, our work serves as evidence that certain "practical" reasons for avoiding FSM-based parsers are invalid.

show abstract

One Server Per City: Using TCP for Very Large SIP Servers

Cited by 13 publications

References 8 publications

Improving authentication performance of distributed SIP proxies

Improving authentication performance of distributed SIP proxies

Overload mitigation mechanism for VoIP networks: a transport layer approach based on resource management

Finite State Machine Parsing for Internet Protocols: Faster Than You Think

Contact Info

Product

Resources

About