Power Saving
Mar 27, 2015
Power Saving
2
Greening of the Internet
• Main idea: Reduce energy consumption in the network by turning off routers (and router components) when they are not being used– Line cards– Crossbar– Main processor
• Why?– Greater deployment possibilities where energy is scarce– Benefits in the event of a disaster
• Question: How much of total power consumption is turning off these components vs. ancillary things like cooling for the data center, etc.– What is the biggest source of overhead?– Micro vs. macro power savings
3
Two Types of Sleeping
• Uncoordinated sleeping– Node makes local decisions about when to sleep– Paper makes some sketchy arguments about packet interarrival
times, etc.
• Coordinated sleeping– Combination of powering off routers and re-routing
4
How and When to Sleep?
• Look at packet interarrival times, etc. and figure out intervals when the line card is not being utilized
• Look at periods of time when the network as a whole is being underutilized– Put parts of the network (i.e., some routers) to sleep when the
whole network doesn’t need to be up and running– What effect does this have on delay, etc.?– How to orchestrate this re-routing?
5
Tight Coupling of Logical and Physical
• Today, the physical and logical configurations of a router is tightly coupled– Hardware upgrade requires logical re-configuration– Customer re-homing requires re-configuration
• The less re-configurations, the better– Less protocol reconvergence– Less traffic disruption– Less configuration errors and overhead
6
VROOM Separates the Logical and Physical
• All logical configurations/states remain the same before/after the migration– IP addresses remain the same– Routing protocol configurations remain the same– Routing-protocol adjacencies stay up– No protocol (BGP/IGP) reconvergence
• Network topology stays intact– Adjacent routers won’t know the router has moved
• Virtually no disruption to traffic– Our most recent results show that the traffic downtime can be
eliminated
7
Application: Power Saving
• Big power consumption of routers– Millions of Routers in the U.S.– Electricity bill: $ hundreds of millions/year
(Source: National Technical Information Service, Department of Commerce, 2000. Figures for 2005 & 2010 are projections.)
1.1
2.4
3.9
0
1
2
3
4
2000 2005 2010
TwH/year
8
Application: Power Saving
• Observation: the diurnal traffic pattern• Idea: contract and expand the physical network according
to the traffic demand
9
Enabling Technologies1. Virtual routers
– Vendors: Cisco VRF, Juniper logical routers, …– Research community: GENI, Cabo, …– Today’s virtual routers have to stay put
2. Live virtual machine migration– Available from VMWare, Xen, …– Typically limited to LANs– Don’t have dedicated forwarding engine (e.g., FIBs, line
cards, …)
10
Enabling Technologies3. Programmable transport layers
– Long-haul links are reconfigurable• Layer 3 point-to-point links are multi-hop at layer 1/2
– Benefit for VROOM: links are easily migratable
Chicago
New York
Washington D.C.
: Multi-service optical switch (e.g., Ciena CoreDirector)
11
Enabling Technologies4. Packet-aware access networks
– Access links are becoming inherently virtualized• Customers connects to provider edge (PE) via pseudo-wires
(virtual circuits)
– Benefit for VROOM: multiple customers can share the same physical interface on PE routers
QuickTime™ and aNone decompressor
are needed to see this picture.
CEPE
CE
TDM Transport Network
QuickTime™ and aNone decompressor
are needed to see this picture.
CEPE
CE
Packet-aware Transport Network
Dedicated physical port
Shared physical port
12
VROOM Architecture
• Virtual link migration– Leverage programmable transport networks
13
VROOM Architecture
• Minimize downtime– In commercial routers, data plane runs in dedicated hardware
(line cards)– Idea: keep the data plane forwarding traffic while migrating the
control plane
14
Deciding Where To Migrate
• Physical constraints– Latency
• E.g, NYC to Washington D.C.: 2 msec
– Link capacity• Enough remaining capacity for extra traffic
– Platform compatibility• Routers from different vendors
– Router capability• E.g., number of ACLs supported
• Good news: these constraints limit the search space of migration destinations
15
VROOM Architecture• Edge migration
– Leverage packet-aware transport networks• Virtualized interfaces (label-based)
• Virtualized access links (pseudo-wires)
• No need for a per customer physical interface on PE routers
16
Events During Migration
• Network failure during migration– The old VR image is not deleted until the migration is confirmed
successful
• Routing messages arrive during the migration of the control plane– BGP: TCP retransmission– OSPF: reliable flooding
17
Prototype Evaluation
• Configure a Xen virtual machine to work as a software router
• Leverage the live migration functionality provided by Xen
• Use GRE tunnels to emulate virtual links– No assumption on the underlying physical/link layer technology
18
Questions
• How does this relate to hot standby?– Relation to MOWER (LAWN gateway) architecture (involves
swapping MAC/IP addresses on two redundant boxes so that upgrades, failures, etc. are transparent to hosts)
• What if all you needed was forwarding state + backup?
• Why process migration as opposed to state migration?– Minimizing the amount of state, etc. can make the migration
easier (?)– Not necessarily, but state that doesn’t change in the events of
failure is easier to migrate (e.g., consider the case of FRR)
19
Questions
• Migrating state vs. migrating processes
• What layer is appropriate for migrating links?– Migrating at layer 2 is inherently limiting– LAWN is a good example: the campus-wide VLAN is stretching
the limits of what is practically feasible (e.g., due to gratuitous use of broadcast, etc.)
– Assumptions about latency, time to recover from failures, etc.
20
Opportunities• DTNs: powering down is a form of disruption• Splicing: Compute backups so that you can power down
without having to do anything special in the control plane• Management and monitoring tools must adapt to account
for “planned failures”• Using designs for planned failures might result in protocols
that are more resilient to unplanned failures• “What-if” scenario evaluation: Where should you power
down? Is it possible to power down Google datacenters (what’s the cost of electricity at that site, at certain times of day, etc.)?
• Power costs of optical vs. copper, etc.• AGORA/Cabo: Spot market for routers, circuits, etc.