OpenFabrics
Software
User Group
Workshop
SC14 Remote I/O Pipeline
Processing Demonstrtion
#OFSUserGroup
Work done under contract to Naval Research Laboratory
Dardo D Kleiner / CIPS, Corp. /
Linden Mercer / PSU /
Demonstration/Test Motivation
JTRS, Wireless, Mobile, etc …
100Gb/s
10Gb/s
1Gb/s
100Mb/s
1Mb/s
Large Numbers of Lite-Clients
Simple Queries Trigger Extensive Bandwidth Utilization
L
A
T
E
N
C
Y
R
E
D
U
C
T
IO
N
Sensors
Super
computers
Distributed
Processing
Distributed
Processing
People
People
Data
Data
Remote I/O Pipeline Processing
Network
100G
Crypto
SCInet
Oakland, CA
NERSC
Chicago
StarLight
New Orleans
SC14
ESNet
100G
Testbed
Controller
Pipelines
Manager
So ware
Defin
e
d
Exchange
Display &
User Interac on
100G
Crypto
X
X
CTL
CTL
NERSC-TB1
STAR-TB1
AOFA-CR5
Source
Data
24 x 4K
Video
(WTN)
Source
Data
24 x 4K
Video
Pipe-lines
Proc-essing
Ce
nt
ur
yL
ink
10
0G
CenturyLin
k
100G
>95% BW
Efficiency
Gigapixels/sec
Network on
demand
100G Type 1 (ESS)
Risk Reduction
Prototype
Processing on
demand
Network/
crypto on
demand
Processing on
demand
Pipe-lines
Proc-essing
Pipe-lines
Proc-essing
Washington, DC
NRL
90 40
10GE
100GE
Testbed
100GE
Testbed
40GE
4x10GE
4x40GE,
12x10GE
2x100GE
100GE
12x10GE,
4x40GE
AOFA-CR5
WASH-CR5
ANA-200G
Europe
ESnet
100G Testbed
(Starlight)
STAR-CR5
CHIC-CR5
MANLAN
ATLA-CR5
SC14 Scinet,
New Orleans
MAX Switch
ESnet
100G Testbed
(NERSC)
NASH-CR5
NERSC-TB1
HOUS-CR5
KANS-CR5
DENV-CR5
NASA
Goddard
NRL demo (100G)
NASA demo
–
Northern path (25G)
NASA demo
–
Southern path (75G)
75 25 25
Caltech demo
–
ANA-200G CERN (100G)
Brian Tierney, ESnet 11/6/2014
SC14 demos
–
ESnet
CERN
100GE
FNAL
40 40Caltech demo
–
NERSC path (100G)
100 100
Starlight
Fabric
AMS
100GE
100VLANs 1802,1810
VLANs 1940-1949
VLANs 2605
VLAN 1860
VLAN 1862
40VLANs 2602,2603
exoGENI Rack
(Starlight)
Circuit SC04 Circuit SC09,10 Circuit SC01ELPA-CR5
75BNL/FNAL demo (40G)
VLAN 2100
University of
Chicago
STAR-TB1
FILENAMESC14-DEMOS-V11.VSD
BNL
SUNN-CR5
PNWG-CR5
10GE
Singapore
CenturyLink
100GE
CenturyLink
100GE
Univ of Chicago demo (40G)
100 40
exoGENI Rack
(OSF)
40GE
25All Links
100G unless
otherwise
noted
exoGENI demo (40G)
VLANs 1921,1922
100
Ciena/Canarie
Research Net
100GE
ANL demo (100G)
VLANs 1925
NEWY-CR5
CenturyLink 100GE to SC14 25 CenturyLink 100GE to SC14 SC03 40NRL
2x10GE
LLNL demo (80G)
Layer 3
Other Demos 10G or Less:
5-7G between PNNL, NERSC, ORNL, ANL, and BNL; DOE Data Replication demo (L3)
10G Loop from U. Chicago to ELPA for UDP/UDT accelerator NIC demo (VLAN: 1911/12)
10G from Singapore to SC14, 5G from Singapore to GA Tech (via ATLA) (VLAN 4000-4004)
10G loop from Aspera Booth at SC14 to Chicago, (VLAN 1930)
90
U Tokyo demo (90G)
VLANs 2987-89
LLNL-CR5
80 Circuit SC02 20 20100G Network Paths
SC14 Pipelines Processing
Detail
Oakland-NERSC or
SC14-NOLA
Chicago or
Wash., DC (NRL)
SC14 – New Orleans, LA
Switch
X RMA targ.
sink src sinkOp onalsrc 40 GbE R0CE Adapter RMA targ. sink src Op onal sink src RMA targ. sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc RMA targ. sink src Op onal sink src sink sink sink sink src sink sink RMA init. sink src 40 GbE R0CE Adapter Switch X RMA targ.
sink src sinkOp onalsrc 40 GbE
R0CE Adapter
RMA targ. sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc RMA targ. sink src
Op onal sink src RMA targ. sink src sinkOp onalsrc RMA targ. sink src Op onal sink src sink sink sink sink src sink sink RMA init. sink src 40 GbE R0CE Adapter Switch X RMA targ.
sink src sinkOp onalsrc 40 GbE R0CE Adapter RMA targ. sink src Op onal sink src RMA targ. sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc RMA targ. sink src Op onal sink src sink sink sink sink src sink sink RMA init. sink src 40 GbE R0CE Adapter Switch X RMA targ.
sink src sinkOp onalsrc 40 GbE
R0CE Adapter
RMA targ. sink src sinkOp onalsrc RMA targ. sink src Op onal sink src RMA targ. sink src Op onal sink src RMA targ. sink src sinkOp onalsrc RMA targ. sink src Op onal sink src sink sink sink sink src sink sink RMA init. sink src 40 GbE R0CE Adapter async i/o
sink src sinkOp onalsrc Adaptec controller async i/o sink src Op onal sink src async i/o sink src sinkOp onalsrc
async i/o sink src sinkOp onalsrc
async i/o sink src
Op onal sink src async i/o sink src sinkOp onalsrc
RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src 40 GbE R0CE Adapter async i/o sink src sinkOp onalsrc Adaptec
controller async i/o sink src sinkOp onalsrc
async i/o sink src Op onal sink src async i/o sink src Op onal sink src async i/o sink src sinkOp onalsrc
async i/o sink src Op onal sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src 40 GbE R0CE Adapter async i/o sink src Op onal sink src Adaptec controller async i/o sink src sinkOp onalsrc
async i/o sink src
Op onal sink src async i/o sink src sinkOp onalsrc
async i/o sink src
Op onal sink src async i/o sink src sinkOp onalsrc
RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src 40 GbE R0CE Adapter async i/o sink src sinkOp onalsrc Adaptec controller async i/o sink src Op onal sink src async i/o sink src sinkOp onalsrc
async i/o sink src sinkOp onalsrc
async i/o sink src sinkOp onalsrc
async i/o sink src Op onal sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src RMA init. sink src 40 GbE R0CE Adapter Switch X 40 GbE R0CE Adapter RMA targ. sink src Op onal sink src RMA targ. sink src Op onal sink src RMA targ. sink src Op onal sink src RMA targ. sink src Op onal sink src sink sink sink src sink 4K pres. sink src NVidia
100G
100G
RMA targ. sink src Op onal sink src 40 GbE R0CE Adapter RMA targ. sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc RMA targ. sink srcOp onal sink src RMA targ. sink src sinkOp onalsrc
RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src 40 GbE R0CE Adapter RMA targ. sink src sinkOp onalsrc 40 GbE R0CE Adapter RMA targ. sink src Op onal sink src RMA targ. sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc RMA targ.
i/o sink src sinkOp onalsrc RMA targ. sink src Op onal sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src 40 GbE R0CE Adapter RMA targ. sink src Op onal sink src 40 GbE R0CE Adapter RMA targ. sink src sinkOp onalsrc RMA targ. sink src Op onal sink src RMA targ. i/o sink src Op onal sink src RMA targ. sink src sinkOp onalsrc
async i/o sink src sinkOp onalsrc
RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. sink src 40 GbE R0CE Adapter RMA targ. sink src sinkOp onalsrc 40 GbE R0CE Adapter RMA targ. sink src Op onal sink src RMA targ. sink src sinkOp onalsrc RMA targ.
i/o sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc RMA targ. sink src sinkOp onalsrc
RMA Init. sink src RMA Init. sink src RMA Init. sink src RMA Init. ini ator sink src RMA Init. sink src RMA Init. sink src 40 GbE R0CE Adapter
Message Bus
Enterprise
Cloud
Sensor
Platform
Ground
Station
Streams
Sink
Transform
Source
Sink
Transform
Source
Sink
Transform
Source
Message Bus
SC14 Pipelines
Processing Detail
Cont’d
async i/o
sink
src
Op onal
sink
src
Adaptec
controller
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
40 GbE
R0CE
Adapter
async i/o
sink
src
Op onal
sink
src
Adaptec
controller
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
40 GbE
R0CE
Adapter
async i/o
sink
src
Op onal
sink
src
Adaptec
controller
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
40 GbE
R0CE
Adapter
async i/o
sink
src
Op onal
sink
src
Adaptec
controller
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
async i/o
sink
src
Op onal
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
RMA init.
sink
src
40 GbE
R0CE
Adapter
100G
Message Bus
Oakland-NERSC and SC14-NOLA
•
EchoStreams FlacheSAN system
– 48 SSD
– 4 x 40 Gbps
•
Sourcing 24 coordinated
uncompressed UHD video
streams
– 4096x1714x24 fps each
– > 4 gigapixels/second (RGB)
– >97 gigabits/second
•
>150 Gbps raw storage to network
performance
•
RDMA networking provides low
demand on CPU – more
processing available for the
mission
SC14 Demo Quicklook Results
disk2null
disk2roce
disk2tcp
null2roce
null2tcp
SC14 Demo Quicklook Results
From 10//2014
disk2roce
0
GB/
s
ec
10
5
0
100
50
Gb
/s
ec
0
50
%
100%
CPU
Band
w
idth
SC14 Demo Quicklook Results
From 11/7/2014
0
100
50
Gb
/s
ec
Band
w
idth
• Early test between NERSC (Oakland, CA) and StarLight (Chicago)
• Single FlacheSAN2 data source
– >23 Gbyte/second raw storage performance
– Network data output of >150Gbps over four 56Gbps InfiniBand or 40Gbps
Ethernet (RDMA).
– 2RU density
– Minimal power consumption of 450W.
• Data streams are “caught” by 4 compact Pipeline processing
systems, each with 40G Ethernet interfaces
100 Gbit Ethernet Type 1
Prototype
EAI Design Type 1 Prototype
• Conforms to NSA Ethernet Security Specification (ESS)
• Available as EE100G Commercial Off the Shelf (COTS) Encryptor
– 100Gb/s AES-256 XPN-MAC-SEC GCM Encryption Core [IEEE
802.1AEbw]
– Verified in Silicon at 400Mhz in Xilinx Virtex-6 HX565T, Virtex-7 2000t
– 512 Security Associations by default (32-64K configurable)
– 128 Internal TCAM Classifier Rules by default (32-16K configurable)
– Highly Configurable Soft/Technology Independent VHDL to support 100GbE
applications in FPGAs or ASICs.
Results
• 4K Video Processing
NRL and Collaborators Conduct 100 Gigabit/Second Remote I/O Demonstration
Participants and Contributors
Naval Research
Laboratory
Orange
Silicon Valley
DREN
#OFSUserGroup
OpenFabrics Software
User Group Workshop
Thank You
NRL
Booth
1639
CenturyLink
Century
-Link
SCinet
ESnet
NRL SC14 Demo
(all VLANs
are “tagged” unless noted)
0
1
0
1
1
1
1
1940,44,45,49
1941,44,46,49
1943,44,48,49
1940,44,45,49
1941,44,46,49
1942,44,47,49
1943,44,48,49
1
2
3
4
NERSC-ECHO
0
1
0
1
1
1
1
1
STAR-ASUS1
0
1
0
1
1
1
1
1
STAR-ASUS2
1942,44,47,49
0
1
0
1
1
1
1
1
STAR-ASUS3
0
1
0
1
1
1
1
1
STAR-ASUS4
NERSC-TB1
STAR-TB1
AOFA-CR5
LAC-OCC-MLXe
0
1
0
1
1
1
1
1940,44,45,49
1941,44,46,49
1943,44,48,49
1940,44,45,49
1941,44,46,49
1942,44,47,49
1943,44,48,49
1
2
3
4
NOLA-ECHO
0
1
0
1
1
1
1
1
NOLA-STEEL1
0
1
0
1
1
1
1
1
NOLA-STEEL2
1942,44,47,49
0
1
0
1
1
1
1
1
NOLA-STEEL3
0
1
0
1
1
1
1
1
NOLA-STEEL4
NOLA-MLNX
1940,44,45,49
1941,44,46,49
1942,44,47,49
1943,44,48,49
1944,49
0
1
0
1
1
1
1
NOLA-TWR
NRL-MLNX
1940,44,45,49
1941,44,46,49
1943,44,48,49
0
1
0
1
1
1
1
1
NRL-SUPER1
0
1
0
1
1
1
1
1
NRL-SUPER2
1942,44,47,49
0
1
0
1
1
1
1
1
NRL-SUPER3
0
1
0
1
1
1
1
1
NRL-SUPER4
NRL-VDX
1940,44,45,49
1941,44,46,49
1942,44,47,49
1943,44,48,49
core-sw
0
1
0
1
1
1
CT
NOLA-CRYPTO
0
1
0
1
1
PT
NRL-CRYPTO
1945-49
1945-49
0
1
0
1
1
1
PT
0
1
0
1
1
CT
Z
Untagged
Z
Untagged
1940-44,Z
1940-49
1940-49
WASH-CR5
1
9
4
0
,4
1
,4
4
,
4
5
,4
6
,4
9
1
9
4
2
,4
3
,4
4
,
4
7
,4
8
,4
9
MAX
10 GbE
40 GbE
100 G
1639-A
1639-F
5/1/
1
1/1/
1
?/?/
?
6/1/1
3/1/
1
1/1/
1
2/1/
1
A – 1940
B – 1941
C – 1942
D – 1943
E – 1944
U – 1945
V – 1946
W – 1947
X – 1948
Y - 1949
11/10/14 11:08 AM