Jereczek, Grzegorz (2017) Software Switching for High Throughput Data Acquisition Networks. PhD thesis, National University of Ireland Maynooth.
Preview
phd-thesis_20170619v4.pdf
Download (2MB) | Preview
Abstract
The bursty many-to-one communication pattern, typical for data acquisition
systems, is particularly demanding for commodity TCP/IP
and Ethernet technologies. The problem arising from this pattern is
widely known in the literature as incast and can be observed as TCP
throughput collapse. It is a result of overloading the switch buffers,
when a specific node in a network requests data from multiple sources.
This will become even more demanding for future upgrades of the experiments
at the Large Hadron Collider at CERN. It is questionable
whether commodity TCP/IP and Ethernet technologies in their current
form will be still able to effectively adapt to bursty traffic without
losing packets due to the scarcity of buffers in the networking hardware.
This thesis provides an analysis of TCP/IP performance in data acquisition
networks and presents a novel approach to incast congestion in
these networks based on software-based packet forwarding.
Our first contribution lies in confirming the strong analogies between
the TCP behaviour in data acquisition and datacenter networks. We
also provide experimental evaluation of different proposals from the
datacenter environment for application in data acquisition to improve
performance and reduce buffer requirements.
The second contribution lies in the design and experimental evaluation
of a data acquisition network that is based on software switches.
Performance has traditionally been the challenge of this approach, but
this situation changes with modern server platforms. High performance
load balancers, proxies, virtual switches and other network functions
can be now implemented in software and not limited to specialised
commercial hardware, thus reducing cost and increasing the flexibility.
We first design and optimise a software-based switch with a dedicated,
throughput-oriented buffering mechanism for data acquisition.
Our experimental results indicate that it performs significantly better
than some typical Ethernet switches under heavy congestion. The
optimised software switch with large packet buffer reaches maximum
bandwidth and completely avoids throughput degradation typical for
hardware switches that suffer from high packet drop counts.
Furthermore, we evaluate the scalability of the system when building
a larger topology of interconnected software switches. We highlight
aspects such as management, costs, port density, load balancing, and
failover. In this context, we discuss the usability of software-defined
networking technologies, Open vSwitch Database and OpenFlow, to
centrally manage and optimise a data acquisition network. We have
built an IP-only parallel leaf-spine network consisting of eight software
switches running on separate physical servers as a demonstrator.
Item Type: | Thesis (PhD) |
---|---|
Keywords: | Software Switching; High Throughput Data Acquisition Networks; |
Academic Unit: | Faculty of Science and Engineering > Research Institutes > Hamilton Institute |
Item ID: | 8851 |
Depositing User: | IR eTheses |
Date Deposited: | 27 Sep 2017 08:53 |
URI: | https://mu.eprints-hosting.org/id/eprint/8851 |
Use Licence: | This item is available under a Creative Commons Attribution Non Commercial Share Alike Licence (CC BY-NC-SA). Details of this licence are available here |
Repository Staff Only (login required)
Downloads
Downloads per month over past year