Energy-efficient Ethernet Networks - Calit2 · Energy Efficient Ethernet Networks UCI – Broadcom...
Transcript of Energy-efficient Ethernet Networks - Calit2 · Energy Efficient Ethernet Networks UCI – Broadcom...
www.broadcom.com
Nicholas IlyadisVice President and Enterprise CTOBroadcom Office of the CTO
Energy Efficient Ethernet Networks
UCI – Broadcom Collaboration
January 12 2010
2
Agenda
Overview of Energy Efficient NetworksMotivation, Savings and End-to-End System Interaction
Energy Efficient Ethernet (802.3az)Standard Objectives, Status and Solutions
EEN – Energy-efficient Ethernet NetworksBroadcom’s End-to-End Philosophy to Energy Savings in Networks
Overview of EEE’s Layer 2 Mechanism and Features
Open Discussion on Collaboration Opportunities
3
Broadcom EEE Vision and Mission
Traditional Networks
Next Generation Networks
Enable end-to-end networks that allow Broadcom’s customers to lead the industry in performance and features in a cost- and
energy-competitive framework
Performance Cost
Energy
4
Industry and Regulatory Trends• Government and Industry Recognition
– April 19, 2006 “Green Grid” formed
– December 20, 2006 House Resolution 5646 signed into law
– European code of conduct
– Japanese government initiative “Top Runner”
• IEEE P802.3az – Energy Efficient Ethernet– Broadcom heavily involved in launching the project in 802.3
• Energy Star– EEE requirements for Servers planned in future draft (2010) once P802.3az is ratified
– EEE requirements for PCs planned in future draft (2010) once P802.3az is ratified
– Historically, EU and other countries will follow suit
– Energy Star has kicked off an enterprise storage elements specification, gaming console etc.
– Specification for SNE (Small Networking Equipment) launched Nov 2009. LNE will be next
• Lower energy usage means lower operating costs
5
Motivation – A Macro View
• All electronics– IT equipment, consumer electronics, telephony
• Residential, commercial, industrial– At least 250 TWh/year– $20 billion/year
• Based on .08$/kWh; rates are rising
– Over 180 million tonsof CO2 per year
• Roughly equivalent to 35 million cars!
• IT equipment about half of this– PCs, displays, printers, servers,
network equipment
One central baseloadpower plant (about 7 TWh/yr)
PCs etc. are digitallynetworked now — Consumer Electronics (CE) will be soon
Numbers represent U.S. only
6
Motivation – A Micro View
Source: Luiz André Barroso, (Google) “The Price of Performance,” ACM Queue, Vol. 2, No. 7, pp. 48-53, September 2005.
(Modified with permission.)
Unrestrained IT power consumption could eclipse hardware costs and put great pressure on affordability, data center infrastructure, and the environment.
8
Motivation – A Link Perspective
• High port count triple speed switches
– Linear relationship of power consumption to number of active links
– Aggregate savings attractive in putting inactive links in LPI
• Low port count 10G systems
– Idle power savings on a single link attractive
Typical switch with 24 ports 10/100/1000 Mb/s
Various computer NICs averaged
Results from 1st order (rough) measurements – all incremental AC power
9
Application – Wiring Closet
EEE EnabledEEE EnabledGig SwitchGig Switch
EEE EnabledEEE EnabledClientsClients
Wiring Closet
Wireless LAN
EEEEnabled GigE
Links
EEEEnabled GigE or 10G
Links (for <100Meter Links)
10
Desktop links have low utilization• Snapshot of a typical 100 Mb Ethernet link
– Shows time versus utilization (trace from Portland State Univ.)
0%
20%
40%
60%
80%
100%
0 1000 2000 3000 4000 5000 6000 7000Time (s)
Util
izat
ion
Typical bursty usage(utilization = 1.0 %)
Fig1.xls
11
Why EEE?
• 48 Port Triple Speed Switch (PHY Only Savings)
– Non-EEE Phy Power = 48*440 mW+ = 21.1W– EEE Phy Power = 48*220 mW = 10.6W – Power Savings on idle switch = 10.6W– 8 High Stack power Savings = 84.5W– 20 Wiring Closets (Large Enterprise) = 1690W– Assuming 100% attached Desktops
= additional 1690W– Total Enterprise Power savings = 3379W
• Assume average traffic is 5% …– $0.10/kWhr * 0.95 * 3.379 KW * 24 hrs
= $7.70 per day– Total Savings = $2,812 per year
+BRCM GPHY power/port
Power Savings Case – Wiring Closet Electricity Costs
100%
50%
Without EEE BRCM EEE PHYsavings*
Lowers Power by50%
*Additional power savings possible with BRCM
12
Application – Data Center and TOR
EEE Gphy EnabledEEE Gphy EnabledServer ControllerServer Controller
Data Center Rack
EEEEnabled 1 and 10 GBASET
Links
EEEEnabled GigE or 10G
Links (for <100Meter Links)
13
Ethernet Traffic Profiles• Snapshot of a File Server with 1 Gb Ethernet link
– Shows time versus utilization (trace from LBNL)
utilization <=1.0 %Start time 12:33 PM 2/8/2007 (30 min)
14
Agenda
Overview of Energy Efficient NetworksMotivation, Savings and End-to-End System Interaction
Energy Efficient Ethernet (802.3az)Standard Objectives, Status and Solutions
EEN – Energy-efficient Ethernet NetworksBroadcom’s End-to-End Philosophy to Energy Savings in Networks
Overview of EEE’s Layer 2 Mechanism and Features
Open Discussion on Collaboration Opportunities
15
IEEE 802.3 (Ethernet) Working Group Organization802.3 (Ethernet)Working Group
Chair: David J. Law
Vice-Chair: Wael William Diab
IEEE 802.3atMDI Power
Enhancements(PoEP)Chair:
Mike McCormack
IEEE 802.3av10G-EPONTask Force
Chair:Glen Kramer
IEEE 802.3azEnergy EfficientEthernet (EEE)
Task ForceChair:
Mike Bennett
IEEE 802.3baHigh Speed
Ethernet 40G/100G Task Force
Chair:John D’Ambrosia
IEEE 802.3TSSG – Time Sync
Study Group
Chair: Steve Carlson
IEEE 802.3Maintenance
Standing Task ForceMaint. Requests &Current Projects:
IEEE 802.3bb,IEEE 802.3bc
Chair: Wael Diab
IEEE 802.3.1Ethernet MIBs
Task Force
Chair:Howard Frazier
Optics Sub-Task Force
Chair: Pete Anslow
Cu Sub-Task Force
Chair: Chris Diminco
Logic Sub-Task Force
Chair: Mark Gustlin
Extended Reach Parallel MMF
Ad HocCo-Chairs: A. Barbieri
John Petrilla
802.3 RulesOwner:
Wael Diab(Vice-Chair)
Layer 2 Ad-Hoc
Chair: Wael Diab
Wake ShrinkageAd-Hoc
Chair: David Law
Layer 2Ad-HocChair:
Wael Diab
Transformerand Channel
Ad-HocChair:
Fred Schindler
16
EEE Objectives• Define a mechanism to reduce power consumption during
periods of low link utilization for the following PHYs– 100BASE-TX (Full Duplex)– 1000BASE-T (Full Duplex)– 10GBASE-T– 10GBASE-KR– 10GBASE-KX4– 1000BASE-KX
• Define a protocol to coordinate transitions to or from a lower level of power consumption
• The link status should not change as a result of the transition• No frames in transit shall be dropped or corrupted during the
transition to and from the lower level of power consumption• The transition time to and from the lower level of power
consumption should be transparent to upper layer protocols and applications
17
EEE Objectives
• Define a 10 megabit PHY with a reduced transmit amplitude requirement such that it shall be fully interoperable with legacy 10BASE-T PHYs over 100 m of Class D (Category 5) or better cabling to enable reduced power implementations
• Any new twisted-pair and/or backplane PHY for EEE shall include legacy compatible auto negotiation
18
Low Power Idle Overview
Low-PowerActive Active
Deassert LPI
Hold
TrTs
Quiet Quiet Quiet
Tq
Wait a minimum of Tw_Sys before sending data (Tw_sys >= Tw_PHY)
Sleep Data / IdleData / Idle Alert Wake IdleRefreshRefresh
Assert LPI
• Low Power Idle (LPI) – PHY powers down during idle periods• During power-down, maintain coefficients and synchronization
to allow rapid return to Active state• Wake times for the respective twisted-pair PHYs
100BASE-TX: Tw_PHY <= 30 usec1000BASE-T: Tw_PHY <= 16.5 usec10GBASE-T: Tw_PHY < ~8 usec (2 modes)
Tw_PHYTw_sys
19
IEEE P802.3az Progress and Timeline
Draft Progress Successful initial Working Group Ballot & Recirc (D2.0 & D2.1)
All comments against D2.0 and D2.1 consideredD2.1 published October 19th, Recirc closed November 4th
D2.2 published December 24th
Recirc closes January 13th
LegendIEEE 802 PlenaryIEEE 802 Interim IEEE-SA Standards Board
Task Force
Sep2007
Jul2007
Nov2007
May2008
Mar2008
Jul2008
Nov2008
May2009
Nov2009
Mar2010
Sep2009
Sep2008
Mar2009
Jan2008
Jan2009
Jul2009
Jan2010
May2010
Jul2010
Sep2010
PAR
TF Review WG Ballot
You arehere
SponsorBallot
Std!
D3.0D2.0D1.0
Last newproposal
Lastfeature
Lasttechnicalchange
Baseline
20
Agenda
Overview of Energy Efficient NetworksMotivation, Savings and End-to-End System Interaction
Energy Efficient Ethernet (802.3az)Standard Objectives, Status and Solutions
EEN – Energy-efficient Ethernet NetworksBroadcom’s End-to-End Philosophy to Energy Savings in Networks
Overview of EEE’s Layer 2 Mechanism and Features
Open Discussion on Collaboration Opportunities
21
Framework• Physical layer
– Rate established at link up– Rate constrains operation of physical layer and upper layers– IEEE P802.3az standard, EEE, allows for suspending the energy on the wire by
turning IDLES to Low Power Idles• Allows energy savings on the PHY layer• Enables further savings up the stack
• MAC Si and Above– Offers a large opportunity for additional power savings– Includes additional interfaces and subsystems– Implementations vary by network position and node architecture– Additional savings enabled by
• Broadcom architectures that leverage PHY EEE• Optimized control policies• 802.3az defines a layer 2 mechanism that allows for additional power savings
22
ControlPolicy
Broadcom’s EEN: End-to-End SavingsEND POINT SWITCH
PHY
MAC
Controller
Controller SW
OS
PHY
MAC
NWSW
NWSW SW
Application
OS
802.3azEEE
ControlPolicy
Coordinatedand Optimized
End-to-End Control Policy
Well Integrated Control Policy and EEE
Within a Box
EEN
23
Broadcom’s EEN: End-to-End SavingsEND POINT SWITCH
PHY
MAC
Controller
Controller SW
OS
PHY
MAC
NWSW
NWSW SW
Application
OSBroadcom Customizable and Optimized Control Policy
• Maximize energy efficiency by maximizing operation in saving states• Minimize performance, latency impact by avoiding unnecessary transitions• Customizable via Software
Broadcom Enabled
Broadcom Enabled
24
Agenda
Overview of Energy Efficient NetworksMotivation, Savings and End-to-End System Interaction
Energy Efficient Ethernet (802.3az)Standard Objectives, Status and Solutions
EEN – Energy-efficient Ethernet NetworksBroadcom’s End-to-End Philosophy to Energy Savings in Networks
Overview of EEE’s Layer 2 Mechanism and Features
Open Discussion on Collaboration Opportunities
25
Layer 1 Support Layer 2 Support
SwitchController
EEE Enhanced Layer 2 Operations
CPU
1000BASE-T MDI
Layer 2Control Policy
• Opportunity to save additional power within a box (link partner)– Additional circuits beyond the PHY can be turned off
• Additional RX wakeup time negotiated using 802.3az’s Layer 2 — Standards based
PCI-e
Server with Controller NIC Switch
26
Review: LLDP Overview
• Operates over a point to point link• Completely enclosed protocol
– We define data, it gets transported• We don't get to make changes to the protocol
• Data in ‘Local MIB’ transported to ‘Remote MIB’– Transported by TLVs (type, length, value)
Local MIB
Remote MIB
Local MIB
Remote MIBA change in a ‘Local MIB’ attribute appears in the corresponding ‘Remote MIB’ attribute
System 1 System 2
Source: diab_02_0109.pdf
27
• Officially called “Data Link Layer” – Effort Chaired by Broadcom in standards• Several Components: (a) Transport (b) State machine (c) MIB (d) Features• In EEE allows for the dynamic negotiation of wake up time for RX
– Allows for cascaded interfaces– Deeper sleep modes– Fallback– Efficient buffering
• In PoEP allows for the dynamic negotiation of power allocation for PD – Power budgeted when it is needed, supplies operate at near optimal efficiency point, power backup
systems and batteries optimized for load (e.g. UPS) – Smart allocation: Features scalability with power budget
• Nearly identical SMs for PoEP and EEE
802.3at’s and 802.3az’s Layer 2
dot3at localsystem MIB
dot3at remotesystem MIB
aRemAbc
aLocDef
State 1
State 2
abc = True
def <= True
def <= False
Chassis IDTLV
Port IDTLV
Time To Live TLV
End Of LLDPDUTLV
dot3atTLV
OptionalTLV
..LLDPDU Format
OptionalTLV
Source: joint_diab_1_0708.pdf
28
Energy Efficient Ethernet TLV
TLV Type = 127
TLV informationString length = 14
802.3 OUI00-12-0F
802.3 Subtype =
TBD
TX Tw_sys
RX Tw_sys
FallbackTw_sys
Echo TX
Tw_sys
Echo RX
Tw_sys
TLV header TLV information string
TLV information string (continued)
7 bits 9 bits 3 octets 1 octet
2 octets 2 octets 2 octets 2 octets 2 octet
...
...
29
Fallback State: Overview
• 2nd “fall back” preference of RX system when requesting Tw_Sys from TX partner
• RX systems have discrete states that TX partner is unaware of.
– TX may have dynamic buffering scheme
• TX can take this information into account when making an allocation decision
• Efficiencies / benefits can be realised • Similar concept exists in P802.3at
30
Agenda
Overview of Energy Efficient NetworksMotivation, Savings and End-to-End System Interaction
Energy Efficient Ethernet (802.3az)Standard Objectives, Status and Solutions
EEN – Energy-efficient Ethernet NetworksBroadcom’s End-to-End Philosophy to Energy Savings in Networks
Overview of EEE’s Layer 2 Mechanism and Features
Open Discussion on Collaboration Opportunities
31
Leading Industry with Next Gen Power Initiative
100%
50%
Without EEE With BRCM EEEPHY
Server Switch
PCStorage
Lowers Powerby
50%