Transcript QoS Support in 802.11 Wireless LANs

Enabling a complementary
connection-oriented internet
Malathi Veeraraghavan
Univ. of Virginia
[email protected]
• Outline
•
•
•
•
Bandwidth sharing in connection-oriented networks
Specific problems
Why do it?
Hasn’t this been attempted before?
Talk at NSF workshop, Santa Barbara, CA, April 12-13, 2005
1
What our optical devices
colleagues have achieved so far
Communications link technology
• optical
Multiplexing technology
• DWDM
Tx/Rx
Communication link  Network
• Switches
Switching modes
Tx/Rx
Switch
optical
switches
Position-based switching
(circuit switching)
 Connection-oriented (CO)
networking mode
Header-data based switching  Continue working
(packet switching)
2
Interdependence
Needs support NOW
Research that leads to
new communications
applications
GMPLS protocols
Research that leads to
networking software
we like packet
switches
Optical circuit switches
Research that leads to
networking hardware
make faster
switching devices
DWDM
Research that leads to
communication link and
switching devices
3
Spectrum of options
Our optical communication technology
colleagues are taking us in this direction
Point-to-point always-up
high-capacity links;
George Glider world of
completely free bandwidth
- no role for networking
Our kids are taking us in the opposite
direction
Dynamically shared
circuits;
Leverage increased
bandwidth offered by
fiber and DWDM;
increased service
quality for some
wasted bandwidth
Bandwidth is
expensive;
Share it to the hilt;
Hence packet
switching
4
Focus of this talk
• Networking “software” research
– Do something useful with the optical
circuit switches created by our
hardware colleagues
• Optical circuit switches implies
– Research on connection-oriented (CO)
mode
– Where should the focus be within this
context?
5
Bandwidth sharing: the very reason for the
existence of networks (switches)
• Research issues in bandwidth sharing in CO
networks?
– Protocols for running the network completely with
software and hardware
• humans just hang out and watch and interfere if something
goes wrong
– Algorithms
• when do you accept a request for bandwidth reservation?
when do you reject? when do you give less than the
requested bandwidth; how much less?
• use of Markov Decision Processes – trade off fairness for
utilization
• how to handle advance reservations for large durations or
large bandwidth in conjunction with small duration or small
bandwith calls
6
Current work
• Protocols for running the CO network
completely with software and
hardware (minimal humans)
– GMPLS protocols
7
GMPLS protocols
• Triumvirate to build large-scale
networks in “plug-and-play” mode
– LMP to discover neighbors
– OSPF-TE for routing
– RSVP-TE for signaling
• Should be able to create distributed
large-scale networks with “minimal”
admin support
8
Amazingly!
• Standards quite advanced
• Implementations exist!
– Switch vendors haven’t waited for
business cases to implement these
protocols
9
Even more amazing!
• Deployed routers and switches now
have the latest versions of software
releases that include GMPLS
protocols
–
–
–
–
Cisco and Juniper routers (MPLS/RSVP)
Sycamore and Ciena SONET switches
Movaz, Calient(?) WDM switches
With Dragon’s help, Ethernet switches
can be run in CO mode
10
But now
• Commercial service providers
– Are waiting for business cases to simply
flick the switch – “turn on” RSVP; “turn
on” MPLS in routers/switches
– Chicken-and-egg problem: apps. and
switch features
• Research network administrators
– Internet2, DoE’s ESnet are “afraid” to
turn these on
11
Need to fund researchers to
• Discover if there’s something to be
“scared” of
• TESTBEDS!
– Look at investments in Canada, Japan,
Europe
– It’s a shame – we are wasting the work
of our optical device colleagues and our
GMPLS engineers
12
Well, hasn’t this happened
before?
• What about our ATM devices/hardware
colleagues’ hard efforts?
• ATM networking software researchers created
the protocols
• Apps were experimented with in universities; trials
were run by service providers
• Didn’t take off. Why?
• Do we give up? Are we saying Connectionless
packet-switched networking is the only way to
build networks?
13
I don’t think so
• Two reasons
– MIPs for processors, MB for memory, ??
for a network service
– Analogy with transportation modes
14
Reason 1
• Purpose of a communication link and by
extension a communication network
– To provide connectivity between a data sending
entity and a data receiving entity
– Quantify connectivity
• bandwidth is a primary measure
– Shouldn’t we have a network that provides
users specific bandwidth levels as and when
requested?
15
Reason 2
• Analogy with people/goods
transportation modes
– unreserved travel: roadways
– reserved travel: airline seat
• So what breaks down in the analogy
for types of networks moving data?
– we only have roadways now
16
So look for ideas from researchers on what
pitfalls are being avoided in their proposals
ATM
Homogeneous CO network
Required new deployment of switches
First, aimed for end-to-end network:
• fiber-to-desktop;
• ATM NICs
Then went for a partial-ATM
solution (Ethernet/IP enterprise
networks + ATM WAN)
• Required egress routers to do
flow classification
• Thinking: if no CO userplane at end host, then no use
for signaling at the end host
Heterogeneous CO internet
Use already deployed switches
(leverage Internet2, ESnet, VLANs)
Allow partial connections
• But have end host signal its
desire to reserve bandwidth
• Avoid all those issues of RSVP
not scaling
• who says all flows need to
reserve bandwidth
17
Partial path reservations
• Ipsilon’s IP switch
Call to make a
reservation
(if only for part of
the distance:
airport-to-airport)
– Flow classification at “airport” to trigger
connection setup
– Questions of scalability – notion of having to
hold “state” information for millions of flows
• No, just the ones who requested bandwidth
CL network
CO network
airport
airport
18
Specific problems
• End-to-end heterogeneous connections
– VLAN capability in Ethernet switches makes this
economically possible
• Partial connections
– Peel off slice of bandwidth on congested link for
connections – for delay-sensitive apps (Internet gaming,
etc.)
• Use MDP to determine Virtual Partitioning limits
for sharing a link between CL service and CO
service
• Billing – fairness/util. algorithms + implementation
19
CO networking research focus
• Then has to center around
– Applications & Testbeds
– Algorithms
– Protocols
– Architectures
– Implementations
that need/enable dynamic sharing of bandwidth
• And have to think large-scale
– The value of the network grows exponentially
with the number of endpoints
20
Summary
• Near-term:
• Fund CO networking research
• Optical circuit switches with GMPLS already here
• Long-term:
• Fund optical switching devices and networking
hardware research to make optical packet
switches, burst switches
• Fund optical comm. technology work to keep
pushing bandwidth higher
• Seems that there’s a capacity glut now but if our kids
have their say, we’ll soon need more
21