The meeting is in Cambridge at the William Gates Building of the computer lab, from 10:30 to 16:30 (room number FW11).
The aim of this meeting is to take a broad overview of the project so far, present, in techincal detail the work done so far and the work planned for the final 22 months of the project. The industrial partners will describe the nature of their work as it relates to the NaaS project. The academic partners will describe their intended work plan for the next stage of the project. The aim is to ensure:
The broad aim of the NaaS project is to investigate in-network line rate processing that can help data centres where their performance is network bound. In particular this involves several elements:
This is an implementation of the first stage of the NaaS vision, an on-path middlebox that can speed up processing for Hadoop and Solr at line rate. This improves the processing power of these technologies and reduces the burden they put on the data centre network.
In this strand of work, the project has built an OpenFlow unikernel. Unikernels are small, type-safe, memory-safe application-specific virtual machines. Our OpenFlow unikernel supports both switch and controller functions, and is built in OCaml, a memory-safe statically-typed language that's immune to certain classes of bug, using Mirage OS. Unikernels are both much smaller and (typically) faster than other virtual machines, e.g., with a memory footprint typically less than 7MB. These OpenFlow unikernels will enable middleboxes with custom sophisticated switching and routing protocols to be managed by a logically centralised controller. At the current time we have an operational OpenFlow 1.4 switch built as a unikernel using the Frenetic libraries.
Netronome is providing products and solutions for accelerating flow-aware packet processing in commodity servers, targeted at vSwitch offload in end-server NICs and accelerating SDN enabled middleboxes. We'll present a quick overview over the products relevant to the NAAS project and highlight some challenges we see in these areas.
Xilinx provides the FPGA that underpins the NetFPGA board. This talk will overview research in Xilinx Labs that has led to the Xilinx SDNet product for doing packet processing at up to 100G rates, driven from a high-level specification. It will also cover current SDN-related research, which includes integrating the control plane of SDNet into current Openflow and next-generation OpenFlow settings. These topics have a close overlap with the NaaS project's aims.
We will summarize Citrix's product interest in software-defined networking and adjacent areas; this will focus on our server virtualization and IaaS cloud products, XenServer and CloudPlatform, but will also touch on our networking products including NetScaler. We will describe XenServer's current and desired future use of relevant technologies such the Open vSwitch and Mirage-based services. We will conclude by sharing our thoughts on the future of network services in a commoditized platform.
BT High Performance Cloud: Analytics as a Service: We will outline the High Performance Cloud services that BT is building to extend our products for the Financial sector. The BT Radianz Cloud community is the world's largest secure networked financial community, consisting of many thousands of financial user locations around the globe. We will specifically focus on Analytics as a Service, as an example of a potential use-case for Open-Mirage. We will close with a summary of the forum BT initiated to standardise the execution environment of virtualised network functions.
Leading organizations worldwide count on NetApp for software, systems and services to manage and store their data. This talk briefly looks at NaaS from the perspective of the storage subsystem.
The project will generalise the NetAgg work to produce a middlebox design that will allow a wide variety of middlebox tasks (both traditional and novel) to be easily aided with in network processing. A specifically designed high-level language will allow developers to simply code the middlebox task appropriately but will restrict the capabilities to ensure that their designs can run at line rate in a high-speed network environment (10G and above). A brief description of the plans for this will be followed by a general discussion about the best ways to proceed.
I describe ongoing and planned work for the development of a hardware platform for in-network computation, based on NetFPGA. Such a platform could host the NaaS box described in the previous talk. This development is intended to lower network latency, improve resource-utilisation, and facilitate scale-out and consolidation. I will describe the main challenges we are facing -- including programming heterogeneous hardware, and moving applications to run on the network -- and how we are tackling them.
After performance tuning our current implementation, our future plans include extending our OpenFlow switch implementation to handle complex application-specific protocols and uses requiring features such as DPI and modification of packet payload data. We are also investigating use of more recent SDN frameworks than OpenFlow such as Protocol Oblivious Forwarding (POF) from Huawei.
This wrap-up session will allow us to critique the progress so far and, more importantly, decide on the next steps for the work.