SOLUTION BRIEF
Ericsson Cloud SDN with Netronome Agilio®
Server Networking Platform Achieves Massive
TCO Savings in Cloud Data Centers
KEY BENEFITS
„„Software-defined
flexiblity and
dynamic control
„„More
data delivered to more
applications per server
„„High-scale
security with high
performance
„„Lower
TCO
Significant challenges arise when
trying to implement Cloud SDN with
high-scale security and performance
purely with a software-based
approach.
Server-based networking is one of the key ingredients of cloud software-defined
networking (SDN) data center deployments on Telco-grade infrastructure such as
the Ericsson HDS 8000 Hyperscale Datacenter System. With Ericsson Cloud SDN,
integration with OpenStack and Open Daylight provides a solid framework for
implementing functions such as dynamic and seamless L3 VPN connectivity. At the
same time, high levels of policy scalability are needed so that distributed cloud infrastructures can be automated and governed for security and compliance. However,
significant challenges arise when trying to implement cloud SDN with high-scale
security purely with a software-based approach. This is because these operations
are extremely CPU intensive, and take away a great deal of server CPU resources that
could otherwise be used to run IT or network functions virtualization (NFV) applications. There is a negative impact on total cost of ownership (TCO) as more servers
are needed to run numerous applications,
Netronome’s Agilio SmartNICs eliminate these problems and improve the TCO
of data centers by providing an alternate way to implement these server-based
networking functions much more efficiently, while at the same time not sacrificing
any of the benefits. Agilio provides the ability to implement the server-based
networking data path directly on the server adapter, offloading the server CPU cores
from needing to perform this function. The acceleration and offload of server-based
networking to Agilio SmartNICs is what allows the same amount of application work
to be performed by far fewer server resources. This improvement in application
work efficiency drives a dramatic improvement in overall TCO for both IT and NFV
workloads.
First, we will look at the IT use case. IT workloads typically are storage or compute
bound, and therefore do not consume as much network data as NFV applications.
Let us assume that for a given IT Cloud deployment, the average application running on a VM or container consumes 300Mbps at an average packet size of 200
Bytes (or about 200K packets/sec). If the server-based networking functions, in this
case driven by Open vSwitch (OVS), are implemented in software using 4 Xeon CPU
cores, there will only be enough data to feed up to 18 applications. In this case there
is extra capacity in the 24-core Xeon server that is effectively stranded and underutilized. The implementation with Agilio, on the other hand, uses only 1 Xeon core
to implement the server-based networking functions, and 23 Xeon cores are available for applications running on VMs or containers. Furthermore, since Agilio can
process and deliver far more network data, it can run many more applications per
SOLUTION BRIEF: Ericsson Cloud SDN with Netronome Agilio Server Networking Platform
Ericsson Cloud SDN with OVS
in User Space and Traditional NIC
Rack Throughput: 72Mpps
Applications per Rack: 360
Ericsson Cloud SDN with OVS
Running on Netronome Agilio Platform
Rack Throughput: 440Mpps
Applications per Rack: 2200
Server Core Allocation
TOR
x86
Core
x86
Core
x86
Core
OVS
4 cores
x86
x86
NIC
Server
NIC
Server
Core
Core
x86
Core
NIC
Server
x86
Core
x86
Core
x86
Core
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
x86
Core
x86
Core
x86
Core
x86
Core
x86
x86
Core
x86
x86
Core
Core or
Core
VMs
containers
x86
x86
20
cores
Core
Core
x86
Core
TOR
x86
Core
x86
Core
x86
Core
x86
Core
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
18 applications at 300Mbps/200Kpps each
NIC
NIC
NIC
Server
Server
TOR
TOR
TOR
TOR
x86
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
x86
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
Core
Core
VMs or
containers
x86
x86
23
cores
Core
Core
x86
Core
x86
Core
110 applications at 300Mbps/200Kpps each
Racks needed to support 2200 VNFs
TOR
Server Core Allocation
OVS
Core
TOR
TOR
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
Server
Racks needed to support 2200 VNFs
TOR
6X
Server
Server
Server
Server
Server
Server
Server
Server
Server
Server
LOWER TCO
Server
Server
Server
Server
Agilio
Figure 1: IT Use Case
server, per rack at the same performance level. The analysis
in Figure 1 shows that it would take over 6 racks to run the
same number of applications when using a traditional NIC
and software OVS, compared to just one rack with Agilio
performing OVS acceleration and offload.
Now we will examine the NFV use case. NFV workloads
typically are lighter on compute and storage, and are
Ericsson Cloud SDN with OVS
in User Space and Traditional NIC
Rack Throughput: 168Mpps
VNFs per Rack: 80
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
Ericsson Cloud SDN with OVS
Running on Netronome Agilio Platform
Rack Throughput: 440Mpps
VNFs per Rack: 220
Server Core Allocation
TOR
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
VMs
8 cores
x86
Core
TOR
x86
Core
x86
Core
OVS
16 cores
x86
Core
x86
Core
optimized to process network data at higher speeds.
Assume that for a given NFV deployment, the average
application running on a VM or container consumes 3Gbps
at an average packet size of 200 Bytes (which equates to
about 2 Million packets/sec). For this workload, with software
OVS, we would have to dedicate many more Xeon cores
to get adequate data delivery to the NFV applications. In
the example below, we dedicate 16 cores to OVS, about the
x86
Core
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
Agilio
Server
4 applications at 3Gbps/2Mpps each
NIC
NIC
NIC
Server
Server
Server
Racks needed to support 220 VNFs
TOR
TOR
TOR
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
NIC
Server
Server Core Allocation
x86
OVS
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
x86
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
x86
Core
Core
Core
VMs or
containers
x86
x86
23 cores
Core
Core
x86
Core
x86
Core
11 applications at 3Gbps/2Mpps each
Racks needed to support 220 VNFs
TOR
Server
Server
Server
Agilio
Figure 2: NFV Use Case
Server
Server
Server
Server
Server
Server
Server
Server
Server
Server
Server
3X
LOWER TCO
SOLUTION BRIEF: Ericsson Cloud SDN with Netronome Agilio™ Server Networking Platform
maximum that can be reasonably allocated due to stability
and scaling issues. Even with so many cores dedicated
to OVS, only enough network data can be processed and
delivered to run 4 NFV applications. The implementation
with Agilio, on the other hand, is able to process and
deliver enough data to run 11 NFV applications at the same
performance level. The analysis in Figure 2 shows that it
would take almost three racks to run the same number of
NFV applications when using a traditional NIC and software
OVS, compared to just one rack with Agilio performing OVS
acceleration and offload.
To explore further workload examples and data center
configurations, and see how your TCO reduction that can be
achieved with Agilio SmartNICs, try out our TCO calculator,
which can be found at www.netronome.com/products/ovs/
roi-calculator.
Summary
This analysis confirms that Agilio SmartNICs can deliver
more data to more applications running on cloud servers
such as the Ericsson HDS 8000, which translates directly to
improved server efficiency and a dramatic reduction in TCO.
This analysis shows a TCO reduction of 6X can be achieved
for typical IT workloads and 3X with NFV workloads. Data
has also been collected for points in between these two
workloads for operators that may have different or mixed
requirements as shown in the following chart:
TCO Savings Using Ericsson Cloud SDN and Netronome Agilio Platform
90
Percent (%) Savings
80
70
60
50
40
30
20
10
0
200 Kpps
400 Kpps
700 Kpps
1200 Kpps 2000 Kpps
Per Applications Workload
Netronome Systems, Inc.
2903 Bunker Hill Lane, Suite 150
Santa Clara, CA 95054
Tel: 408.496.0022 | Fax: 408.586.0002
www.netronome.com
©2017 Netronome. All rights reserved. Netronome, the Netronome logo, and Agilio are trademarks or registered trademarks of Netronome
Systems, Inc. All other trademarks mentioned are registered trademarks or trademarks of their respective owners in the United States and
other countries.
SB-ERICSSON-CLOUDSDN-1/17