ETHERNET ROADMAP UPDATE Scott Kipp skipp@brocade.com October 2014 1
DISCLAIMER The views expressed in this presentation are the views of the presenter and not of the Ethernet Alliance. 2
Ethernet Roadmap Updates Since the August meeting, several standards are being worked on in IEEE 802.3 25GbE Study Group met for the first time 25GBASE-T Call for Interest (CFI) was announced Next Generation Enterprise Access BASE-T PHY CFI announced If these pass, we could have 5 new speeds being developed simultaneously: 1. 2.5GbE 2. 5GbE 3. 25GbE 4. 50GbE 5. 400GbE 10/10/2014 3
>1B Ports Shipped in 2014! Over 7 Billion Ethernet ports shipped in last decade! Ethernet Switch Ports 1 Router Ports 2 Controllers and Adapters 3 PoE Devices 4 Access Ports 5 Total 2004 195M 1.2M 110M* 8.9M 62M 377M 2014 452M 1.8M 294M 98M 129M 974M Probably another 100M ports on widgets, TVs, video games and other equipment in 2014. With Internet of things and vehicular applications expected to ship 100s of Millions of ports by the end of the decade, Ethernet will surpass 1.5B Ports/year by 2020. 1. Dell Oro Ethernet Switch Layer 2+3 Report, July 2014 2. Dell Oro Routers Report, July 2014 3. Dell Oro Controller and Adapter Report, January 2012 4. Based on PoE switch ports from Dell Oro Ethernet Switch Layer 2+3 Report, July 2014 5. Dell Oro Access Five Year Forecast, July 2014 * 2006 values since 2004 values weren t available
Ethernet Alliance Roadmap Find more at: www.ethernetalliance.org/subcommittees/roadmapping-subcommittee/ Name Speed Date Initial Standard Ratified 10Mb/s Ethernet 10 Mb/s 1983 100Mb/s Ethernet 100Mb/s 1995 Gigabit Ethernet 1 Gb/s 1998 10 Gigabit Ethernet 10 Gb/s 2002 25 Gigabit Ethernet 25Gb/s 2016 (est)* 40 Gigabit Ethernet 40 Gb/s 2010 100 Gigabit Ethernet 100 Gb/s 2010 400 Gigabit Ethernet 400 Gb/s 2017 (est.)** *Estimated on a 2-year standardization process that started with the CFI in July 2014 **Estimated on a 4-year standardization process that started with the CFI in March 2013
Speed (b/s) Ethernet Speeds Log 1T 400G 100G 100GbE 400GbE Key: Ethernet Speed Speed in Development 40G 40GbE 50GbE 10G 10GbE 25GbE 5GbE 1G GbE 2.5GbE 100M 100 Mb/s Ethernet 1990 2000 2010 2020 Initial Standard Completed
Why 2.5G to 50G? Installed base of CAT5E cabling driving 2.5GBASE-T 10GBASE-T doesn t support 100m of CAT5E Installed base, new technology and low cost of SFP+ driving 25GbE Serial lanes lower cost than parallel lanes of 40GbE 40GBASE-T looks very challenging, but 25GBASE-T looks easier 100GbE still prohibitively expensive for many users 1X40GbE SFP+ should deliver good bandwidth/$ compared with 40GbE QSFP+ (4X10G) Can 50GbE SFP+ be deliver better bandwidth/$ too?
Figure from Upcoming 2.5G CFI 10/10/2014 8
Why 25G Now? Best cost/gbps soon Technology is ready ASICs are port limited, so need more Gb/s/port 640Gb/s 64 Ports of 10G 5X More 3.2 Tb/s 128 Ports of 25G 64 10GbE port ASIC enables 48 SFP+ and 4 QSFP+ 640Gb/s of Throughput 128 25GbE port ASIC enables 32 QSFP+ 3.2 Tb/s of Throughput
5X The Calories McDonald s Hamburger Burger King s Triple Whopper with Cheese -1250 calories -250 calories
1000X The Storage 2003 512MB 2014 512GB! 10/10/2014 11
How Long are Bits and Frames? A 1GFC Frame is about 4km long and a 1GFC bit is 7.41 long A 32GFC Frame is about 125 m and a 32GFC bit is about ¼ long Data Rate Bit time speed of light in glass bit length bit length frame length bits/second picos/bit m/s m/bit inches/bit m/frame 1GFC 1,062,500,000 941 200,000,000 0.188 7.41 3975.53 2GFC 2,125,000,000 471 200,000,000 0.094 3.71 1987.76 4GFC 4,250,000,000 235 200,000,000 0.047 1.85 993.88 8GFC 8,500,000,000 118 200,000,000 0.024 0.93 496.94 16GFC 14,025,000,000 71 200,000,000 0.014 0.56 248.47 32GFC 28,050,000,000 36 200,000,000 0.007 0.28 124.24 100fs My Master s Thesis at LLNL 0.1 300,000,000 0.000030 0.001181 10/10/2014 12
25GbE Standards Current 25GbE Study Group has following objectives: Backplanes like 802.3bj 3m of Twinax 5m of Twinax 100m like 100GBASE-SR4 25GBASE-T CFI coming next month Don t know what distance or cabling yet? 40GBASE-T defined for 30m over CAT8 in letter ballot www.ethernetalliance.org 10/10/2014 13
Electrical Lane Speed (b/s) Electrical Interface Speeds When will the next speeds arrive? 100G 10G XFP MSA standardizes XFI 10Gb/s in 2005 802.3bj standardizing CR4 4X25Gb/s in 2014 1X40G 2.5X/9 Years = 11%/year nx100g 8X50G Gigabit Ethernet (no electrical standard for 1 Gb/s) in 1998 10X/7 Years = 39%/year 1G 100M Fast Ethernet standardizes 100 Mb/s in 1995 10X/3 Years = 116%/year Key: Ethernet Speed Future Speed 1990 2000 2010 2020 2030 IEEE Standard Completed
Electrical Lane Speed (b/s) 11% CAGR is Fairly Slow 100G 50G 40G 25G 400GbE Task Force voted to do 8X50G electrical by 2017 100GBASE-CR4 in 2014 40G Serial in 2018? 8X50G in 2020? 100G in 2027 XFP MSA standardizes XFI 10Gb/s in 2005 2.5X/9 Years = 11%/year Key: Ethernet Speed Future Speed 10G 2000 2010 2020 2030 Standard Completed
Electrical Lane Speed (b/s) 20% CAGR is Great! 100G 100G in 2020 50G 40G 25G 100GBASE-CR4 in 2014 8X50G in 2017? 40G Serial in 2016? 10G SFP+ MSA standardizes SFI 10Gb/s in 2009 2000 2010 2020 2030 Standard Completed 2.5X/5 Years = 20%/year Key: Ethernet Speed Future Speed
How far will 40G Serial go? Bandwidth Length Product depends on Media Speed Reach with PCB Loss =10dB MMF Reach (OM3) BASE-T Cable Distance 1G 20-36 550m 100m on CAT5 10G 10-15 300m 100m on CAT6A 25G 4-6 100m?? 40G?? 70m? 30m on CAT8 50G??? 50m??? One example of an Ethernet Link PHY PCB 1 SFP+ SFP+ PCB PHY PCB Loss 1 Media Loss PCB Loss 2
It s Only Getting Harder Higher speeds equal higher loss and power or shorter distances CAUI-4 (4X25G) Chip to Module Insertion Loss 10dB @ 25Gb/s (12.89GHz) 50Gb/s is off the chart unless advanced modulation is used (12.89GHz with PAM-4) When will we get 100Gb/s signaling? >20dB @ 40Gb/s (20GHz)
Where s the Debate? Should we standardize 40GbE Serial in SFP+? Should we standardize 50GbE Serial in SFP+? Should we standardize 100GbE Serial in SFP+? 4X100GbE QSFP112 should combine to 400GbE 400GbE has many options: Should 8X50G Electrical Interface be NRZ or PAM-4 Should optics be 8X50G or 4X100G l and what encoding (NRZ, PAM-4, DMT, CAP, QAM)? Many BASE-T questions to be discussed later
Modules Sold / Year 10G Recap SFP+ and QSFP+ Win! Before 25GbE and 50GbE Came Along 100M 2015 SFP+ surpasses 10M! 10M 2010 SFP+ Becomes the Clear Winner 1M 100K XFP and Others never reach critical mass 2016 40G QSFP+ >1M 2000 2005 2010 2015 2020 Based on LightCounting Forecasts
Modules Sold / Year How does 25GbE Affect this? 100M New 10GbE and 40GbE forecasts reduced because of 25GbE and 100gbE 10GbE SFP+ 10M 25GbE SFP28 1M 100K 40GbE QSFP+ 2005 2010 2015 2020 2025 Based on LightCounting Forecasts until 2016 100GbE QSFP28
Modules Sold / Year 40G Will SFP+ win? 100M 10M 40GbE QSFP+ 1M 100K 2010 2015 2020 2025 QSFP+ Up to 2016 Based on LightCounting Forecasts Will 40G SFP40 surpasses 1M? How about 50GbE SFP50?
Modules Sold / Year 100G QSFP28 then SFP+? 100GbE Challenges 10M 100G QSFP28 expected to be winner 1 2 1M 100K Scale Shifted to 10K! 10K CFP and others never produce >1M/year 100G SFP+ Takes off When? 2005 2010 2015 2020 2025 Based on General Trends
Challenging the Industry The Ethernet Alliance has created two challenges related to 100GbE: 1. The Holy Cup of 100GbE Lambda 2. The Holy Grail of 100GbE SFP+ See Details at: http://www.ethernetalliance.org/subcommitt ees/roadmap-subcommittee/
Modules Sold / Year 400G Still Developing 10M 1M Will 2 nd Gen 400GbE (8X50G) CFP2 become the Clear Winner? Does 3 rd Gen (4X100G) 400GbE QSFP100 happen within 10 years? 100K Scale Shifted to 10K! 10K 2010 2015 2020 2025 2030 Based on General Trends 1 st Gen (16X25G) 400GbE CDP, CDFP, embedded optics and others?
Speed (b/s) Ethernet Speeds 2010-2025 1T 400G 100G 40G 25G 10G 1G 100GbE 10X10G 40GbE 4X10G 4x10.3125G -XLAUI 10X10.3125G CAUI-10, CPPI 100GbE 4X25G CAUI-4 4x25.8G 400GbE 16X25G 25GbE XXVAUI 1x25.8G CDAUI-16 16x25.8G 400GbE 8X50G CDAUI-8 8X50G 40GbE XLAUI-1 1X40G 2010 2015 2020 2025 Technology Availability 1.6TbE 16X100G 50GbE LAUI-1 1X50G TbE 10X100G 400GbE 4X100G 100GbE CAUI-1 nx100g Key: Ethernet Speeds Ethernet Electrical Interfaces Future Speeds Stretched Symbols = Time Tolerance
Where s the Debate? What modulation should be used for 50Gb/s? NRZ and PAM-4 both could work, but which will dominate? Where should the 50G Electrical be used? 2 nd Gen 400GbE (8X50G) is a given Should we do 3 rd Gen 100GbE (2X50G)? Should we have 2 nd Gen 50GbE (1X50G)? 1 st Gen 50GbE (2X25GbE) is in 25G Ethernet Consortium When will 100G Optics be used? 4X100GbE being debated in 400GbE (802.3bs) now 100GbE QSFP28 Single Lambda not far away 100GbE SFP+ is out a ways
The Rate Debate is Next Week! Where: Santa Clara Convention Center When: Next Thursday, October 16 th http://www.ethernetalliance.org/the-rate-debate/ 10/10/2014 28
Morning Agenda Time Topic Speaker(s) Ethernet Alliance Roadmap 9:10am Intro Networking Data Rates inside 9:30am the Data Centers Server Diversity: Why One 9:50am Speed Is No Longer Enough Implications of the Next 10:10am Signaling Rate on Ethernet Speeds 10:30am Lane Rates 10:50am Break 11:05am Discussion on Sessions 11:35am Guest Speaker Scott Kipp, Ethernet Alliance President (Brocade) Alan Weckel, Dell Oro Group Dave Chalupsky, Intel Kapil Shrikhande, Dell Rob Stone, Broadcom Scott Kipp, Ethernet Alliance President (Brocade) Gavin Cato, Dell
Afternoon Agenda Time Topic Speaker(s) 1:00pm Data Center Fabrics Thomas Scheibe, Cisco 1:25pm Introduction of Session Scott Kipp, Ethernet Alliance President (Brocade) 1:30pm Faster 56Gb/s Standardization David Stauffer, Kandou Bus Efforts at the OIF Nathan Tracy, TE Connectivity 1:50pm The Impact of Ethernet Rates on Optics Mitch Fields, Avago 2:10pm 50Gb/s: The Next Unifying Per Lane Rate Chris Cole, Finisar 2:30pm Discussion on Session Scott Kipp, Ethernet Alliance President (Brocade) 2:55pm Break 3:15pm BASE-T Panel: 2.5 to 25G Dave Chalupsky, Intel 4:00pm Flexible Ethernet Rates for Router Stephen Trowbridge, Alcatel Lucent to Transport 5:00pm Wrap-up John D Ambrosia, Dell 5:30pm Networking Reception All 10/10/2014 30
THANK YOU 31