1. Next Generation Spine & Core Data Center Switches

2. Virtual Chassis Fabric
Access Switches
IP Fabric
MPLS Fabric
MC-LAG
Network Overlays
VxLAN
Virtual Chassis Fabric
Flexibility of architectures
10GbE/40GbE access and spine solutions
Software innovations

3. Simple Ethernet Fabric
Spine Switches
IP Fabric
MPLS Fabric
MC-LAG
Network Overlays
VxLAN
Simple Ethernet Fabric ?
High density 40GbE and 100GbE Supporting 10GbE & 25GbE transitions in access
Evolving Cloud data center architectures (VxLAN, MPLS)
Platform for continued data growth

4. Introducing QFX10000 Series Spine Switches
FIXED & MODULAR 10G / 40G / 100G SPINE / CORE SWITCHES
Powered by Juniper
custom silicon
Meet rapid and continuing data growth
MOST SCALABLE
Accelerate innovation
OPEN
Invest for today and tomorrow
FUTURE PROOF

5. Fixed QFX10000 series switches

6. QFX Q
Fixed platform to support transition from 10GbE to 40GbE and 100GbE
Compact form factor with high density

7. QFX Q
Fixed platform to support transition from 10GbE to 40GbE and 100GbE
Half the port density of QFX Q for smaller spine configuration

8. QFX10002-72Q and QFX10002-36Q Front to back cooled system
Resilient power design for mission critical applications

9. 100GbE Support 100GbE SUMMARY
QFX Q
QFX Q
Within a group of 3 ports , one of the port can be used as 100GbE port (SR4, LR4)
The ports marked in red become disabled if 100GbE is in use within a port group (3 ports make up the port group)
All ports can be 40GbE
All ports can be 4X10GbE
100GbE SUMMARY

10. Port Numbering Conventions
set chassis fpc <slot> pic <pic-slot> {port|port-range} <0..71> channel-speed {10g|40g}
et – 40GbE/100GbE
xe – 10GbE ( xe 0/0/0:0-3)
GbE – 1GbE

11. QFX10002 system architecture
Fabric Chip
QFX10002 system architecture
Port ASIC
QFX Q
QFX Q
Multi Chip System to scale I/O and support flexible applications
Each PFE provides 12X40GbE , 4X100GbE OR 48X10GbE interfaces
6 PFEs for QFX Q and 3 PFEs for QFX Q
2 Fabric ASICs for QFX Q and single fabric asic for QFX Q
SYSTEM ARCHITECTURE SUMMARY

12. PACKET BUFFERING SUMMARY
QFX10002 Packet Buffering
Packet Buffer
QFX Q
QFX Q
Each PFE Chip in QFX10000 has 4GB of external packet memory to accept packets as they ingress the system
Optimized for system use is variety of applications including Data Center Edge
Differentiates the system from “Switch on Chip” Systems where packet memory is internal to the PFE
Provides good performance for applications without ECN (Explicit Congestion Notification) or DC-TCP
PACKET BUFFERING SUMMARY

13. QFX10002 Table Memory TABLE MEMORY SUMMARY
Packet Buffer
Table Memory
QFX Q
QFX Q
Each PFE Chip has dedicated memory for storing forwarding tables (256K IPv4 FIB, 256IPv6 FIB, 2M host routes)
Allows collapsing of multiple layers of networking in data centers (DCI and DC Core/Spine)
Small on chip memory per PFE for MAC , ACL and MPLS tables
TABLE MEMORY SUMMARY

14. PFE ASIC PFE SUMMARY 500G Chip ( 500G in, 500G out)
4X100GbE interfaces when providing 10GbE , 40GbE and 100GbE
5X100GbE interfaces when providing 40GbE or 100GbE interfaces
PFE SUMMARY

15. Hybrid Memory Cube
High bandwidth, energy efficient, high density memory
Critical to building a high I/O system
10.2x the bandwidth of DDR3 module
8.5x the bandwidth of DDR4 module

16. System Scale FEATURE QFX10002-36Q QFX10002-72Q Mac Addresses 256K 512K
L2 Domains
16K
L3 VPN Scale 4K
Filters
64K match conditions
8K attachment points
Policers & Counters
MPLS
64K
FIB
256K IPv4, 256K IPv6
Host Scale 2M
VxLAN
32K

17. Fabric ASIC
Crossbar fabric that provides PFE-to-PFE connectivity using CCL (Chip to Chip Links) using 25GbE Serdes
144 X 25GbE SERDES
Uses 1XP protocol
Cell based Fabric ( more details to follow)
SUMMARY

18. PFE to FABRIC Connectivity
12 X 40GbE
4X100GbE
48X10GbE
Fabric
Fabric Chip
PFE
Port ASIC
25Gbps
400Gbps

19. Cell Based Forwarding SUMMARY
Fabric
Fabric
Fabric Chip
Cells are forwarded on all available fabric links.
Port ASIC
On Fabric interface, packet is split into N cells of variable length.
Cells are received and reassembled into a packet
PFE
PFE
PFE
PFE
PFE
PFE
Packet arrives on input port and buffered.
Input Port
Output Port
Page is sent across the fabric (8KB or 16KB)
Cell sizes can vary between 96B, 112B, 128B, 144B, 160B and 176B
SUMMARY

20. Virtual Output Queueing
Problem Statement (Egress buffering)
Fabric
Port1 1
Ingress
Port3
Egress
Port2 2

21. Virtual Output Queueing
Head of Line Blocking with Egress buffering only
Fabric
Port3
Port1
Back Pressure
Ingress
Port3
Egress

22. Virtual Output Queuing
Fabric
Port1 1
Ingress
Port3
Egress
Port2 2
Ingress
VoQ Buffering

23. Benefits of VoQ
Queueing for each egress port maintained on ingress PFE
One place to queue the traffic
Traffic not sent over the fabric if egress port not ready to transmit the traffic resulting in very high system utilization
Prevents Head of Line Blocking

24. VoQ Virtual Output Queues Fabric Fabric PFE PFE PFE PFE PFE PFE
Packet Buffer
Table Memory
PFE
4GB
PFE
PFE
PFE
PFE
PFE
Virtual Output Queues

25. VoQ Architecture for 72X40GbE Ports
4GB Buffer
VOQ P1
P72
Each PFE has 8 queues X 72 ports for a 72X40GbE system
576 Virtual Output Queues per PFE
3,456 Virtual Output Queues per QFX Q System running in 72X40GbE mode
Per Port buffer 57MB
VoQ STRUCTURE

26. VoQ Architecture for 72X40GbE Ports
Virtual Output Queues
In QFX Q System with 72X40GbE ports

27. Virtual Output Queuing
SUMMARY
Virtual Output Queue refers to a queue on output port that is maintained by the input port
Each PFE maintains 8 Virtual Output queues for all ports
VOQs can be of varying size
“Use-meters” to determine buffer allocation fairness
72X40GbE system
3456 Virtual Output Queues
288X10GbE system
13824 Virtual Output Queues
24X100GbE system
1152 Virtual Output Queues
Mixed mode system : 56 X 40GbE + 4 X 100GbE
2880 Virtual Output Queues

28. Packet Forwarding Pipeline
500 Mpps packet processing/lookup capacity to accommodate for multiple lookups within the pipeline
333Mpps packet throughput capacity
PACKET PIPELINE

29. QFX10002-72Q and QFX10002-36Q Control Plane
SUMMARY
4-core Intel Gladden Ivy-Bridge CPU
16GB, 1333MHz of un-buffered ECC DDR3 SDRAM
Field-upgradable BIOS
10G connection between CPU and each PFE
Two Slim SATA Gen 2 SSD slots on board with 32GB NAND Flash module
Provides a reset push-button*

30. Modular QFX10000 series switches

31. Modular QFX10000 Series Switches
6Tbps/Slot
21 RU
Width
17.4”
Depth
35”
6Tbps/Slot
480 X 100GbE
576 X 40GbE
16 I/O Slots
13 RU
8 I/O Slots
Width
17.4”
Depth
32”
240 X 100GbE
288 X 40GbE

32. QFX10008 SUMMARY Dimensions: Height = 22.55” (13 RU) Width = 17.4”
6Tbps/Slot
SUMMARY
Dimensions:
Height = 22.55” (13 RU)
Width = 17.4”
Depth = 32”
Orthogonal Direct Connectors
No midplane between Line Cards and Fabric Cards
8 Linecards/ IO slots
2 Control Boards / Route Engines
6 SIBs/ Fabric cards
2 Fan Trays
2 Fan Controller Boards
6 Power Supplies
Airflow: Front-to-Back
13 RU
Depth
32”
Width
17.4”
288 X 40GbE
240 X 100GbE

33. QFX10016 SUMMARY 6Tbps/Slot 21 RU Depth 32” Width 17.4” 576 X 40GbE
Dimensions:
Height = 36.65” (21 RU)
Width = 17.4”
Depth = 35”
Orthogonal Direct Connectors
No midplane between Line Cards and Fabric Cards
16 Linecards/ IO slots
2 Control Boards / Route Engines
6 SIBs/ Fabric cards
2 Fan Trays
2 Fan Controller Boards
10 Power Supplies
2850W AC or DC 220V input Cable type C19
220V input
Feed redundancy N+1
Airflow: Front-to-Back
Operational Temperature = 6000
21 RU
Depth
32”
Width
17.4”
576 X 40GbE
480 X 100GbE

34. Detail Overview – Rear View
Add the 16 slot
QFX10008
QFX10016

35. Midplane Less Design Power and Cooling efficiency Reliability
Future Scale

36. Modular Line Card Options
36X40GbE with QSFP+
12X100GbE with QSFP28
144X10GbE with 40 to 4X10GbE Breakout
30X100GbE with QSFP28
24X40GbE + 6X100GbE

37. 144X10GbE with 40 to 4X10GbE Breakout
100GbE Ports on 36 Port Card 1 5 29 25 23 19 17 13 11 7 35 31 4 2 6 10 8 12 16 14 18 22 20 24 28 26 30 34 32 3 9 15 21 27 33
100GbE ports
36X40GbE with QSFP+
12X100GbE with QSFP28
144X10GbE with 40 to 4X10GbE Breakout

38. 84X10GbE with breakout cables
10GbE Card
Same 100GE stuff for the 10GE line card
60X10GbE + 6QSFP+
2 X100GbE QSFP28
84X10GbE with breakout cables

39. 84X10GbE with breakout cables
10GbE Card
Same 100GE stuff for the 10GE line card
100GbE
60X10GbE + 6QSFP+
2 X100GbE QSFP28
84X10GbE with breakout cables

40. 36X40GbE Card Fabric Fabric Fabric Fabric Fabric Fabric PFE PFE PFE
Packet Buffer
Table Memory
Fabric
Fabric
Fabric
Fabric
Fabric
Fabric
6X25G
PFE
4GB
PFE
4GB
PFE
4GB
12 X 40GbE
4X100GbE
48X10GbE
12 X 40GbE
4X100GbE
48X10GbE
12 X 40GbE
4X100GbE
48X10GbE

41. 30X100GbE Card Fabric Fabric Fabric Fabric Fabric Fabric PFE PFE PFE
Packet Buffer
Table Memory
Fabric
Fabric
Fabric
Fabric
Fabric
Fabric
6X25G
PFE
2GB
PFE
2GB
5X100GbE
5X40GbE
PFE
2GB
5X100GbE
5X40GbE
PFE
2GB
5X100GbE
5X40GbE
PFE
2GB
5X100GbE
5X40GbE
PFE
2GB
5X100GbE
5X40GbE
5X100GbE
5X40GbE

42. 60X10GbE Line Card Fabric Fabric Fabric Fabric Fabric Fabric PFE PFE
Packet Buffer
Table Memory
Fabric
Fabric
Fabric
Fabric
Fabric
Fabric
6X25G
PFE
4GB
PFE
4GB
48X10GbE
48X1GbE
12X10GbE+6X40GbE
12X10GbE+2X100GbE

43. System Scale QFX1000 (8 slot & 16 slot) FEATURE Mac Addresses 1M
L2 Domains
16K
L3 VPN Scale 4K
Filters
64K match conditions
8K attachment points
Policers & Counters
MPLS
64K
FIB
256K IPv4, 256K IPv6
Host Scale 2M
VxLAN
32K

44. Control Board for Modular QFX10000 switches
SUMMARY
CPU: Intel Ivy Bridge Gladden 4-core 2.4 GHz
Memory:
8GB onboard NAND Flash, 16GB SDRAM
2 removable SATA SSD slots, 1 USB 2.0
4 SFP+ ports
PTP:
RJ45 for Grandmaster clock
1 SFP or RJ45 management port, 1 console port
Same RE / CB in 8 and 16 I/O slot systems

45. Software Architecture
CLI, XML, Netconf, UNIX and API Access
OPEN APIs
PFE
PLATFORM
CARRIER-CLASS
NETWORKING
EXTENSIBLE USER SPACE
3rd party applications
HARDWARE ABSTRACTION LAYER
KVM
KVM
Yocto Linux
CARRIER-GRADE
LINUX
DATA PLANE
x86 CONTROL PLANE
FLASH MEMORY (UBOOT + ONIE)

46. Software Architecture
SOFTWARE ARCHITECTURE HIGHLIGHTS
Decouple Platform & PFE Daemons from Junos
Accelerate TTM for product variants (use Linux drivers & abstractions for new platforms)
Improve performance – Multicore CPU
Improve BFD performance
Mac learning (by taking two big processes out of JUNOs)
Convergence
Direct access to PFE via APIs (Future)
Pace of innovation

47. Optics Options for Data Center
DAC (1-5m)
4X10GbE LR(10Km)
4X10GbE IR(2Km)
4X10GbE SR (300m)
10GbE
40GbE LX4 (150m)
40GbE SR4 (300m)
40GbE ESR4(400m)
40GbE LR4(10Km)
40GbE IR4(2Km)
DAC (1m-7m)
40GbE
100GbE SR4(100m)
100GbE LR4(10Km)
100GbE CWDM4 (2Km)
100GbE
Add lengths

48. What is Junos Fusion?
Data center networking with simplified management at scale
Fusion
Open Standards & programmability
IEEE 802.1BR and JSON-RPC APIs
Resilient
Plug-and-play provisioning
1GbE-100GbE

49. Thank You