全光 IP 网：神话还是现实？——Juniper

1、luc ceuppens vice president, product marketing infrastructure product group 2 copyright 2010 juniper networks, i outline optical bypass: what, why and how? the forgotten piggy challenges myths and realities conclusion and summary 3 copyright 2010 juniper networks, i optical bypass: an introduction n

2、etwork planning technology where optical connectivity is topologically richer than ip/mpls node map. ip/mpls nodes are by-passed by select wavelengths inside the transport optical element (optical bypass) otn circuits (opaque bypass, otn bypass) often popularized as a tool to reduce network capex an

3、d energy footprint reduction of transit traffic (and hence ports) on ip/mpls node on shorter segments (less than 1,500km) a “bypass” wavelength can be optically amplified in the node without oeo conversion, further reducing the cost in essence, the promise is that bypass shifts the capex budget out

4、of ip/mpls equipment towards optical gear. 4 copyright 2010 juniper networks, i optical bypass: good idea? ob is backed up with significant amount of research data and whitepapers. several peer-reviewed capex network model studies were done to prove ob monetary/energy savings. list of popular titles

5、 include “evolution of wdm optical ip: cost and energy” (r. tucker) “packet layer topologies of cost optimized transport” (nsn) deliverables within european “ist nobel 2” project and others several service providers (swisscom, dt, ft) have endorsed optical bypass in presentations and articles in opt

6、ical-themed magazines. ob also enjoys significant analyst coverage. note 1: ob is sometimes referred to as “optical circuit switching” ocs, or “optical cross-connect” oxc) note 2: this presentation uses several graphics from above mentioned papers 5 copyright 2010 juniper networks, i components of o

7、ptical bypass 1. wdm/roadm shelves (in red) serving ip/mpls nodes 2. selective wavelength routing in the optical layer 6 copyright 2010 juniper networks, i optical bypass: only the beginning? transition from all-electrical switching (“opaque”) in every node to ob/ocs (“transparent”) is often viewed

8、as precursor to all-optical packet processing: in this “stage 1”, ocs elements are often shown operating under g-mpls control for dynamic path provisioning 7 copyright 2010 juniper networks, i optical bypass: elevator pitch 1. ip/mpls port cost (in abstract units) 2. oxc/roadm port cost (in abstract

9、 units) replace packet switching by optical switching = lower cost! 8 copyright 2010 juniper networks, i optical bypass: elevator pitch (contd) did anyone notice a pink piggy? if not, lets try again 9 copyright 2010 juniper networks, i optical bypass: the forgotten piggy there is a “pink piggy” in t

10、he network cost picture, so large that people barely notice it. it is the same piggy that drove the transformation from circuit to packet as well as ip convergence. this piggy is called “cost of bandwidth”. it feeds upon available bandwidth and returns monetary services. it heavily favors statistica

11、l multiplexing over circuit switching because it can eat much less and monetize much more. so, why do some people ignore it? 10 copyright 2010 juniper networks, i optical bypass: challenges the smallest unit in ocs is a “wavelength”. optical bypass is contradictory to service providers desire to imp

12、rove spectral efficiency. the higher the wavelength speed, the less opportunity for bypass literature actually acknowledges this traffic matrices tell a different story than bandwidth matrices dynamic provisioning of a light path thru a network of optical switches (e.g. “packet over light lsp”) requ

13、ires complex power/amplification computations for oxc nodes and can take several minutes 11 copyright 2010 juniper networks, i optical bypass: myth 1 ip/mpls nodes in the core are routing packets the packet route is already known at the core ingress point. however, packet core switches are not doing

14、 routing they are intelligent statistical multiplexors filling the core bandwidth at packet-level resolution. opportunity to replace core routers by lower cost lsrs edge 12 corecoreinner core copyright 2010 juniper networks, inc. optical bypass: myth 1 ip/mpls nodes in the core are routing packets p

15、acket network evolution other networks other networks 13 copyright 2010 juniper networks, i optical bypass: myth 2 transit traffic can be efficiently “removed” 14 optical bypass: myth 2 transit traffic can be efficiently “removed” traffic in real networks follows complex distribution patterns and st

16、atic de-multiplexing wastes bandwidth traffic bc overload reduced peak rate x/2 x/2 overload overload traffic ac bandwidth x optical split option (b) unutilized unutilized unutilized unutilized unutilized unutilized unutilized unutilized x x copyright 2010 juniper networks, inc. unutilized optical s

17、plit option (a) 15 copyright 2010 juniper networks, i optical bypass: myth 3 optical bypass saves ip/mpls ports by off-loading transit traffic, the need for ip/mpls ports is reduced wdm network operates at wavelength speeds. a router that needs to talk to n peers over fully transparent network needs

18、 n ports with capacity equal to lambda speed number of ip/mpls ports grows! doctored traffic volumes and low inter-node connectivity degree are the mandatory conditions to demonstrate reduction in requirements for ip/mpls ports 16 copyright 2010 juniper networks, i optical bypass: myth 4 “transparen

19、t” network scale to complex topologies claim: all-optical cross-connect network can link many packet nodes over dedicated point-to-point light paths full-mesh in optical domain requires n*(n-1)/2 light paths, where n is the number of nodes. for wdm density of 80 channels, networks as small as 10-15

20、nodes can easily run of out of channels cost would be prohibitive! complexity of edge devices increases exponentially 17 failed copyright 2010 juniper networks, inc. unusable unusable optical bypass: myth 5 “transparent” networks offer good resiliency claim: all-optical cross-connect network offers

21、high availability if a light path that crosses several nodes has one broken section, the entire path becomes unusable. oxc network can only offer 1:1 protection, effectively doubling provisioned bandwidth. dynamic control plane (g-mpls) cannot help because there are no traffic sources suitable to fi

22、ll the surviving sections unusable 18 copyright 2010 juniper networks, i so, what is the verdict? “optical bypass” offers minimal traffic processing at the cost of inefficient bandwidth utilization. this is a valid design compromise in the following cases: optical spectrum is free (dark fibers or la

23、mbdas) there are intermediate nodes with very small tributary traffic network designers considering ob should remember the piggy! cost of bandwidth can not be ignored statistical multiplexing maximizes bandwidth utilization installing expensive packet nodes is justified by the cost savings in other network resources such as link capacity. our tests show that it pays off to install pack