naas.github.io/tgres.html at master · NaaS/naas.github.io · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
<!DOCTYPE html>
<html>
  <head>
    <meta charset='utf-8'>
    <meta http-equiv="X-UA-Compatible" content="chrome=1">

    <link rel="stylesheet" type="text/css" href="stylesheets/plain.css" media="all" />
    <title>Network-as-a-Service project</title>
  </head>

<body class="html not-front not-logged-in no-sidebars page-node page-node- page-node-58 node-type-project" >
  <div id="skip-link">
    <a href="#main-content" class="element-invisible element-focusable">Skip to main content</a>
  </div>
    <div id="wrap">
    <div class="container">
        <!-- #header -->
                <div id="header" class="sixteen columns clearfix">

            <div class="inner">


                                <h1>Network-as-a-Service (NaaS) project </h1>
                                <div id="name-and-slogan" class="element-invisible">

                                        <div id="site-name" class="element-invisible">
                    <a href="/" title="Home" rel="home">Network-as-a-Service (NaaS) project</a>
                    </div>


                </div>
                            </div>
        </div><!-- /#header -->

<!-- #navigation -->

        <div id="navigation" class="sixteen columns clearfix">

            <div class="menu-header">
                            				<div class="content">

<ul class="menu"><li class="first leaf"><a href="/">Home</a></li>
<li class="leaf"><a href="/news.html">News/Events</a></li>
<li class="leaf"><a href="/people.html">People</a></li>
<li class="leaf"><a href="/publications.html" title="">Publications</a></li>
</ul>
</div>
</div>
</div>
<!-- /#navigation -->

                  <div class="region region-content">
    <div id="block-system-main" class="block block-system">


  <div class="content">
    <div class="panel-flexible panels-flexible-24 clearfix" >
<div class="panel-flexible-inside panels-flexible-24-inside">
<div class="panels-flexible-region panels-flexible-region-24-center panels-flexible-region-first panels-flexible-region-last">
  <div class="inside panels-flexible-region-inside panels-flexible-region-24-center-inside panels-flexible-region-inside-first panels-flexible-region-inside-last">

<div class="panel-separator"></div><div class="panel-pane pane-entity-field pane-node-body"  >


  <div class="pane-content">
    <div class="field field-name-body field-type-text-with-summary field-label-hidden"><div class="field-items"><div class="field-item even" property="content:encoded"></div>
<h2>NaaS technical advisory group meeting 10th December 2014 </h2>

<h3>Attendees</h3>
<ul>
<li>Cambridge: Marco Forconesi, Thomas Gazagnaire, Hwanju Kim, Anil Madhavapeddy, Andrew Moore, Nik Sultana
</li><li>Imperial College: Abdul Alim, Richard Clegg, Paolo Costa, Peter Pietzuch, Lukas Rupprecht, Alex Wolf
</li><li>Nottingham: Masoud Koleini, Derek McAuley, Richard Mortier, Carlos Oviedo
</li><li>BT: Bob Briscoe, Winston Carrera
</li><li>Citrix: James Bulpin, Rob Hoes, Dave Scott
</li><li>NetApp: Lars Eggert
</li><li>Netronome: Rolf Neugebauer, Stuart Wray
</li><li>Xilinx: Gordon Brebner
</li>
</ul>

<h3>NaaS project overview (Peter Pietzuch)</h3>

<p><a href="tg1pres/prp-naas-techgroup-cambs-14.pptx">Powerpoint slideset</a></p>

<h3>Progress to date -- NetAgg box (Paolo Costa)</h3>

<p><a href="tg1pres/pc-naas-techgroup-cambs-14.pptx">Powerpoint slideset</a> Discussion points -- "Why are you making cleverness in the network when the trend is to
dumb network, clever endpoints?"  "East-West style networks might change what you are doing?"  "Different topologies
(e.g. toroidal) can give more flexible options"  "Incast remains a problem that NaaS can solve."
</p>

<h3>Progress to date -- Mirage/OpenFlow (Masoud Koleini)</h3>

<p><a href="tg1pres/mrc-naas-techgroup-cambs-14.pptx">Powerpoint slideset</a>.
Discussion points -- "What about switching latency?" "OpenFlow is not 'high-level' -- app developers don't want to know at this level."</p>

<h3>Industrial Partner presentations: Netronome (Rolf Neugebauer)</h3>

<p><a href="tg1pres/rn-naas-techgroup-cambs-14.pdf">PDF slideset</a>
Discussion points: "Is the host machine aware of the network co-processor?"  "Yes in a limited way."  "At 10G forwarding with flow state
uses many cores as you get > 10,000 flows in ovs".  "'How do you steer packets into the VMs?' 'Could do DMA from NIC.'" Reference to <a href="http://www.ece.cmu.edu/~ece739/papers/maple.pdf">MAPLE paper</a></p>

<h3>Industrial Partner presentations: Xilinx (Gordon Brebner)</h3>

<p> <a href="tg1pres/gb-naas-techgroup.pdf">PDF slideset</a> Discussion points: "New OpenFlow versions coming out regularly, moving target for implementation." </p>

<h3>Industrial Partner presentations: Citrix (James Bulpin)</h3>

<p><a href="tg1pres/jb-naas-techgroup-cambs-14.pdf">PDF slideset</a>
 Discussion points: "People don't expect to pay for a hypervisor, providers of such are diversifying.
Foundations for selling other services and products."  "'Flexible platform for application means need way for applications
to express their needs?' 'Provide basic network abstraction, network processing pushed to application, simple networking layer.'" "'What sort of throughput for Netscaler SDX?' 'Can max out 8x10G ports, polling not interrupt handling.'"
</p>

<h3>Industrial Partner presentations: BT (Winston Carrera, Bob Briscoe)</h3>

<p> Discussion points: Network-as-a-Service vision "Medium" latency (lower than commodity but not "lowest
possible") as service for those where lat totally critical (e.g. finance).  "Analytics as a Service"
analyse data on the wire in real time.  APIs expose product catalogue of services that could be composed
together -- APIs communicate via SDN controller to implement on pre-built network carving up slices of
network management.  Customers can provision their own clients to slice the available bandwidth.
Agile optical network.  Customers can specify the ability to burst over their agreed data rates automatically
and be billed extra.  Packets stored for 20 minutes and triggered conditions cause longer term storage for
diagnostics.  Data acquisition architecture is vital (robust one in place).  How small can we make VM
infrastructure we use?  Suggestion xen/mirage might be useful in this space.
</p>

<p> Discussion points: Network function virtualisation (NFV) standardisation ongoing -- working very fast
for a standards body.  Largest ever ETSI ISG.  Open Platform OPNFV launched Oct 2014 -- Open Source as
fast path to standardisation.  "CAPEX and OPEX reduced but does it enable anything new?"  Enables faster
deployment, shorter window for development.
</p>

<h3>Industrial Partner presentations: NetApp (Lars Eggert)</h3>

<p><a href="tg1pres/naas-netapp.pdf">PDF slideset</a>
Discussion points: Manage workloads with Service Level Objectives is important.  Storage
hierarchy deepening -- keep data off network where possible.  "Can't have petabytes of memory
so something needs to communicate across network".
"Who is programmer for NaaS?  What is their skill set?"  "BT model has same challenge, nets guys
understand OSPF.  Those running DCs best placed to think about these problems."  "Users for NaaS
are 'application author' AND 'datacentre manager'."
</p>

<h3>NaaS box, application specific middlebox (Richard Clegg)</h3>

<p><a href="tg1pres/rgc-naas-techgroup-cambs-14.pptx">Powerpoint slideset</a>
Discussion notes: "What about encryption?  Problem with terminating encrypted streams
at a middlebox."  "What data structures should language allow?  Associative arrays?"
</p>

<h3>Endpoint teleportation (Nik Sultana)</h3>

<p><a href="tg1pres/ns-naas-techgroup-cambs-14-v5.pptx">Powerpoint slideset</a>
</p>

<h3>Mirage SDN switching and control (Richard Mortier)</h3>

<p><a href="tg1pres/mort-naas-techgroup-cambs-14.pptx">Powerpoint slideset</a>
<a href="tg1pres/mort-naas-techgroup-cambs-14.pdf">PDF slideset</a>
Discussion notes: "PoF and P4 are working on a merged solution."  "'Why is x86 faster than
ARM?' 'Possibly issues with memory mapping, tuning may improve results for ARM'"
"Are you assuming a single manager for the infrastructure?"
"Show Xen no worse for fundamental latencies.  Scheduling decisions across hardware boundaries."
"Orchestration across entire orchestra is complex."
</p>

<h3>Progress summary, next steps (Alex Wolf)</h3>

<p>Project is work in progress, bridging layers int he stack, different levels of
abstraction.
Themes: Is it appropriate to be talking about middleboxes or should
we be talking about virtual switches?  Is end-to-end principle dead
or are we happy for applications to interact more closely with network?
Standards are an open question?

Universities should be involved with setting standards?
Perhaps research should lead and be ahead of standardisation?

Can you solve "network" problem in isolation impacts on storage +
computation?  Security remains an issue.

Middlebox versus end-to-end: Applications spread across many entities,
not "traditional middlebox" -- analogous to building applications on
AWS (PaaS).  This is another part of disaggregation story not a
middlebox.

Middlebox implementation detail of intelligent network -- hard problem
is knowing how to split application between network and servers when
both are processing elements.  Middlebox perhaps wrong as a term?

What things will become industry trends?  Zoom out "is netagg different
from just changing toplogy?"  Is general purpose computer sufficient for most
needs -- trend to homogeneity?

Key driver must be cost -- in terms of developer time, cost of hardware
resources and management.  Diverse hosts with different properties
"low latency", "big storage", "many network ports" could prove a
more cost-effective solution.

NetFPGA part of a family of devices that do similar things?

Need to find a differentiator for the project "Don't want to be 'just
another middlebox'"

TAPS group for standardisation may be relevant to project.

How "clean slate" do we want to be as a project?  Do we target legacy
applications or are we "bolder" and assume datacentres could change if
a new approach was sufficiently beneficial?

Regardless of technology/code, could the lessons learned be passed on?

Core assumption is "cooperative environment" -- realistic?  In a cloud
environment application may not "play ball".  Often clients like
"constant performance" -- not going to understand "something got slower
because something else got 'hot'"?

"Maple: Simplifying SDN Programming Using Algorithmic Policies"
http://www.ece.cmu.edu/~ece739/papers/maple.pdf

Stuart has P4 parser written in python -- would like to Open Source --
could be useful ingredient.  Gordon also interested in P4 convergence
and could provide help/advice on NetFPGA.

If customers have flexibility to put code onto devices this needs
a lot of care.  Some customers are not willing to let people know about
source -- want opaque binary blob due to proprietary code.

Most likely meeting time would be in one year.
</p>

</div></div></div>  </div>


  </div>


  </div>
</div>
</div>
</div>
  </div>
</div>
  </div>

            </div>

        </div><!-- /#content -->


        <div class="clear"></div>


	</div>

	<div id="footer" >
        <div class="container">
        	<div class="sixteen columns clearfix">

                <div class="one_third">
                                </div>

                <div class="one_third">
                                </div>

                <div class="one_third last">
                                </div>

                <div class="clear"></div>


                <div class="clear"></div>

        	</div>
        </div>
    </div>

</div> <!-- /#wrap -->
  </body>
</html>