White Paper
Intel
®
IXP23XX Product Line of
Network Processors
Performance of the Intel
®
IXP23XX
Product Line of Network Processors for
Access and Edge Network Applications
www.intel.com/go/
networkprocessors
Contents
Introduction
............................................................................................................................................................................ 2
Product line overview ....................................................................................................................................................... 2
Performance metrics
............................................................................................................................................................... 3
Performance assessment of the Intel
®
IXP2350 network processor
........................................................................................ 3
ATM RAN application example ......................................................................................................................................... 4
Additional application examples ...................................................................................................................................... 5
Conclusion
.............................................................................................................................................................................. 5
Introduction
Manufacturers of access and edge equipment have long
sought the ideal balance of performance and programming
flexibility in a processing platform. The fact is, applications
such as ATM-based Radio Access Networks, DSL Access
Multiplexers (DSLAMs), Wireless Local Area Network (WLAN)
access points, WAN multi-service switches, and others, must
support a range of protocols and line rates to meet application
requirements. Standards-based interfaces and reusable code
can help ensure reliable performance as well as accelerate
time-to-market and leverage software assets across multiple
solutions. And for equipment manufacturers to maximize
profitability they must choose the right network processor—
one that is neither over-powered nor under-powered, yet offers
plenty of headroom in which to innovate.
Intel offers an ideal solution with the Intel
®
IXP23XX product
line of network processors, designed to enable a broad range
of access and edge applications. Built on the same hardware
and software architecture as the entire Intel
®
IXP2XXX product
line, the IXP23XX network processors support deep packet
inspection, traffic management, and forwarding at up to 2
Gbps line rates in a single chip, while providing the headroom
developers need to deliver value-added network services.
each network processor are available with Intel XScale
®
core
operating frequencies ranging from 600 MHz to 1200 MHz,
and fast path microengine operating frequencies ranging
from 300 MHz to 900 MHz. Both network processors are
fully programmable, yet have the processing power to run
applications that previously required expensive, high-speed
ASICs and external control plane processors.
As with all network processors in the entire Intel IXP2XXX
product line, the IXP23XX network processors execute data
plane tasks in parallel across individual microengines. The
Intel
®
Internet Exchange Architecture (Intel
®
IXA) Software
Framework complements the IXP2XXX modular hardware
architecture by providing tested software building blocks, called
microblocks, that developers can use to quickly assemble
functional sequences, or pipelines, that perform specific
network functions. These microblocks are highly reusable
across applications; for example, an IP forwarding microblock
could be used in a DSLAM application with an ATM media
interface, as well as in a WAN multi-service switch.
In this white paper, Intel presents a performance assessment of
the Intel IXP2350 network processor for a variety of common
access and edge applications. In the application examples
discussed, packets are processed in microblocks and
executed across the microengines within the IXP2350 network
processor, processing a comprehensive set of functions
for a variety of data rates and communications protocols.
As the performance assessment will show, the IXP2350
network processor offers robust performance with substantial
headroom even under the most demanding conditions.
Product line overview
The IXP23XX product line includes the Intel
®
IXP2325
network processor, which supports up to 2.5 giga-operations
per second with two 32-bit independent multi-threaded
microengines, and the Intel
®
IXP2350 network processor,
which supports up to 4.9 giga-operations per second with four
32-bit independent multi-threaded microengines. Variants of
2
Performance metrics and IP
forwarding example
When examining the performance capabilities of a network
processor, it is important to first consider the performance
metrics used. Two of the primary performance metrics used in
the communications and networking industry are packet rate
and data rate (see Table 1).
Data rate is the total number of Megabits-per-second (Mbps) that a network-
ing element can handle. Packet rate is the number of million-packets-per-sec
(Mpps) that the networking element can handle. The relation between the
data rate and packet rate is simple:
Data rate (Mbps)
Packet rate (Mpps) = -----------------------
8
x
Packet size (bytes)
and carry different levels of overhead. Therefore, for any
given data rate, performance—measured in ‘million packets
per second’—will ultimately be a factor of packet size and
overhead. With maximum data rates of 3.6 Gbps in the
IPv4 forwarding example in Table 2, it is evident that the
percentage of headroom in the IXP2350 network processor
increases as the packet size grows.
Performance assessment of the
Intel IXP2350 network processor
®
In assessing the performance of the IXP2350 network
processor, Intel chose to use minimum packet sizes in all
application examples. By taking this approach, the IXP2350
network processor is subjected to the most demanding
conditions likely to be encountered in the real world. Even in
these “worst case” scenarios, the IXP2350 network processor
demonstrates robust performance with ample headroom in
which equipment vendors can add value.
To determine performance capabilities for the IXP2350
network processor, Intel calculated the headroom available
with all four microengines processing minimum-sized packets
with a robust set of application functions at a given data
rate. These calculations are based on data from an Intel
IXP23XX cycle and data accurate simulator and on previously
measured data for the Intel
®
IXP2400 network processor
extrapolated to match the specifications of the IXP2350
network processor.
Ethernet Packet Size
Intel
®
IXP2350 Network Processor
• Intel XScale
®
core – 900 MHz
• Microengines (ME) – 900 MHz
64 B
128 B
512 B
1518 B
25% ME headroom (5.35 Mpps**)
58% ME headroom (3.04 Mpps**)
75% ME headroom (0.85 Mpps**)
78% ME headroom (0.30 Mpps**)
Table 1:
Packet Rates vs. Data Rates
In network processors, the unit of work is a packet and,
therefore, packet size is a critical parameter in characterizing
performance. For example, smaller packets will require a
greater number of packets per second to be processed at
a given data rate. Since a certain amount of overhead is
associated with each packet (e.g., table lookups, header
updates, enqueues/dequeues), a greater number of smaller
packets being processed per second will mean more work
for the network processor and, consequently, less headroom
available for value-added features. Conversely, with a fewer
number of larger packets at the same data rate, the network
processor can complete the same task with less work, thereby
retaining greater headroom to support value-added features.
A simple application such as IP forwarding for Ethernet
packets illustrates this point. An IP forwarding application can
be implemented in three stages: packet receive, IP forwarding
based on longest prefix match and packet transmit. Table
2 presents estimated performance and packet processing
headroom for the Intel
®
IXP2350 network processor in
this application.
Communications protocols, such as Gigabit Ethernet and
OC-12, each support different minimum-sized packets
Table 2:
IPv4 Forwarding over Ethernet Performance
** Based on maximum bus frequency and associated data rate of 3.6 Gbps
on the media interface.
3
ATM RAN application example
The ATM RAN application is intended for transport line cards
in the 3 GPP UTRAN network. In this application, the Intel
®
IXP2350 network processor is tasked with processing packets
from the line interface to the backplane and out a second line
interface. Figures 1 and 2 illustrate the full set of functions
performed by the microengines within the IXP2350 network
processor for the ATM RAN application.
In this performance assessment, two configurations of the
line interface are analyzed: 2xOC-3, which is best suited for
the Iub interface of the RNC, and 16 T1/E1, which is best
suited for the Iub interface of Node B. In both configurations,
½½½½½½½½½
½½½½½½½½½½½½½½½
½
½½½
½½½½½½½
½½½½½½½½
½½
½½½½½½½½½½
½½½½
½½½
½½½½
½½½½
½½½½½½½½½½½½
½½½½½½
½½½½½½½½½½½½
½½½½½½½½½½
½½½½½½½
½½½½½½
½½½½½½½½½½½½½
½½½½½½½½
½½½½½½½½½
½½½½½½½½
½½½
½½½½½½½½½½
½½½½½½½
½½½½½½½½½½
½½½½½½½½
½½½½½½½½
½½½
½½½½
½½½½
½½½
½½½½
½½½½½½
½½½
½½½½½
½½½½½½
½½½½½½½½½½½½
½½½½½½½½½
½½½½½½½½½½
½½½½½½½½½½½½½½½
½½½½½½½½½½½½
½½½½½½½½½½½½½½
½½½½½½½½
½½½
Figure 1:
Data flow from the line interface to backplane. These software blocks are implemented on the microengines of the IXP2350 network processor.
½½½½½½
½½½½½½½½½
½½½
½½½½½½½
½½½½
½½½½½½
½½½½
½½½
½½½½½
½½½½½½½½½½
½½½½½½½½½½½½
½½½½½½½½½
½½½
½½½½½½½½½½
½½½½½½½½½½½
½½½½½½½½½½
½½½½½
½½½½½½
½½
½½½½½½½
½½½½½½½½½½½
½½½½½½½½½½½½½½½
½½½½½½
½½½½½½½½½½½½½
½½½
½½½½½½½½½½
½½½½½½½½½
½½½½½½
½½½½½½½½
½½½½½½½½
½½½½
½½½
½½½½½½½
½½½½½½½½½½
½½½½½½½½
½½½½½½½
½½½
½½½½½
½½½½½½
½½½½½½½½½½½½
½½½½½½½½
½½½
½½½½½½½½½½½½
½½½½½½½½½½½½½½
Figure 2:
Data flow from the backplane to the line interface. These software blocks are implemented on the microengines of the IXP2350 network processor.
4
评论