It is commonly accepted that decoupling network functions (NFs) from the underlying physical infrastructure using virtualization and cloud technologies enables service innovation. By means of Software Defined Networking (SDN), the programmability of the Network Function Virtualization (NFV) infrastructure has gained a huge potential for supporting the deployment of NFs in a variety of (virtualized) environments, including Internet and cloud service providers, campus and enterprise networks, and over-the-top applications.

One of the fundamental questions is: Are the software-based solutions (ever going to be) able to cope with the forever changing and increasing traffic demands? Network Function Performance Analyzer is a publicly available, open-source measurement application, which is not only in accordance with standardized methodologies (RFC 2544), but also makes possible to comprehensively compare performance metrics of NFs in an exhaustive range of dimensions. More precisely, NFPA answers

• how a software-based NF
• implemented in a generic language (e.g., C/C++)
• running on a generic platform (e.g., Intel Xeon)
• over a generic operating system (e.g., Linux)
• in different environments (e.g., virtual machine)
• using different drivers performs
• under different traffic patterns.

Complying with RFC 2544, our NFPA is a standalone benchmarking tool, connected to the Device Under Test (DUT), as depicted in the figure above. The NFPA user connects to an NFPA node, i.e., a computer that is able to run NFPA and is connected directly to a separate computer running the desired network function. It's typically connected via two interfaces, however, it is possible to use only one bidirectional link. Then, the user determines the measurement target on the NFPA node by all its relevant parameters, e.g., details of the hardware and software components, number of repeated measurements and their duration, and selects the traffic traces to be used. Afterwards, NFPA sends packets on port 0 (and also on port 1 in case of bidirectional measurements), while receives (a portion of those) packets on port 1 (and also on port 0 respectively), then it calculates the throughput of the DUT in terms of packet/s and bit/s. Once the measurement is finished, NFPA analyzes the results and plots them in the preset units, and in parallel saves the measurement data in a local database.

In addition, we provide a central server with a web-based API, whereby the NFPA user can upload and share the results for further comparison, and in return the user can browse the set of results measured by others.

NFPA’s engine is implemented in Python and relies on standard libraries. It can be configured both by config files and via a lightweight web GUI. In order to avoid the limitation of kernel space network card drivers, NFPA’s network interface is built on Intel’s DataPlane Development Kit (DPDK) for fast packet processing: in particular, for sending and receiving traffic, NFPA uses PktGen with custom Lua scripts for parameterizing, automating and controlling the measurements. In order to support portability, the results are stored in a local SQLite database. Result charts are created using Gnuplot. The overall set of measurements at our central node is stored in PostgreSQL database.

With the increased number of measurement needs and users, new traffic traces have been published, where in most cases the header fields have been randomized. Why is it important to use and match for random header fields? The answer is not trivial, but we have measured that in some cases network functions, such as Open vSwitch, are able to use prefix-es for flow rules making it to be more efficient than it might be in a more realistic scenario (for more details, see how caching (micro and megaflow cache) is implemented in Open vSwitch).

Nevertheless, to ease the usage of these pcap files for your network functions, we extended our site with a Use-cases page, where more details can be found.

Name	#diff. flows	Description	Packet Size (bytes)								Use-case
			64	128	256	512	1024	1280	1500	other
trL2_1	1	Different destination MAC addresses									l2-switch	(483M)
trL2_10	10	Different destination MAC addresses
trL2_100	100	Different destination MAC addresses
trL2_1000	1000	Different destination MAC addresses
trL2_10000	10.000	Different destination MAC addresses
trL2_100000	100.000	Different destination MAC addresses
trL3_1	1	Different destination IP addresses									l3-router	(484M)
trL3_10	10	Different destination IP addresses
trL3_100	100	Different destination IP addresses
trL3_1000	1000	Different destination IP addresses
trL3_10000	10.000	Different destination IP addresses
trL3_100000	100.000	Different destination IP addresses
trPR_12	12	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port									l2-switch l3-router	(500M)
trPR_24	24	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port
trPR_36	36	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port
trPR_48	48	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port
trPR_100	100	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port
trPR_1000	1000	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port
trPR_10000	10.000	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port
trPR_100000	100.000	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port
trPR_500000	500.000	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port									l2-switch l3-router	(2.3G)
trPR_1000000	1.000.000	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port									l2-switch l3-router	(4.5G)
trPR_2000000	2.000.000	Different dst and src MAC addresses, different src and dst IP addresses, different src and dst port									l2-switch l3-router	(8.9G)
LB_ACL_1	1	L2 headers are the same, random src IP, dst IP address=192.168.2.1, random src port and 80 dst port									load_balancer	(489M)
LB_ACL_10	10	L2 headers are the same, random src IP with random src port, 70% of the flows have random dst IP addresses are random from range 192.168.2.1 - .10 with dst port 80
LB_ACL_100	100	L2 headers are the same, random src IP with random src port, 70% of the flows have random dst IP addresses are random from range 192.168.2.1 - .100 with dst port 80
LB_ACL_1000	1000	L2 headers are the same, random src IP with random src port, 70% of the flows have random dst IP addresses are random from range 192.168.2.1 - .100 with dst port 80
LB_ACL_10000	10.000	L2 headers are the same, random src IP with random src port, 70% of the flows have random dst IP addresses are random from range 192.168.2.1 - .100 with dst port 80
LB_ACL_100000	100.000	L2 headers are the same, random src IP with random src port, 70% of the flows have random dst IP addresses are random from range 192.168.2.1 - .100 with dst port 80
AGW_1flows	1	L2 headers are the same, random local src IP (10.x), random dst IP address, random src port and dst port									agw	(16M)
AGW_10flows	10	L2 headers are the same, random local src IP (10.x), random dst IP address, random src port and dst port
AGW_100flows	100	L2 headers are the same, random local src IP (10.x), random dst IP address, random src port and dst port
AGW_1000flows	1000	L2 headers are the same, random local src IP (10.x), random dst IP address, random src port and dst port
AGW_10000flows	10.000	L2 headers are the same, random local src IP (10.x), random dst IP address, random src port and dst port
AGW_100000flows	100.000	L2 headers are the same, random local src IP (10.x), random dst IP address, random src port and dst port
trPR_v2_12	24	Different src MAC addresses, same dst MAC address, different src and dst IP addresses (but exluding LAN, localhost, and Multicast addresses, different src and dst port									l3-router_v2	(399M)
trPR_v2_24	12	Different src MAC addresses, same dst MAC address, different src and dst IP addresses (but exluding LAN, localhost, and Multicast addresses, different src and dst port
trPR_v2_36	36	Different src MAC addresses, same dst MAC address, different src and dst IP addresses (but exluding LAN, localhost, and Multicast addresses, different src and dst port
trPR_v2_48	48	Different src MAC addresses, same dst MAC address, different src and dst IP addresses (but exluding LAN, localhost, and Multicast addresses, different src and dst port
trPR_v2_100	100	Different src MAC addresses, same dst MAC address, different src and dst IP addresses (but exluding LAN, localhost, and Multicast addresses, different src and dst port
trPR_v2_1000	1000	Different src MAC addresses, same dst MAC address, different src and dst IP addresses (but exluding LAN, localhost, and Multicast addresses, different src and dst port
trPR_v2_10000	10.000	Different src MAC addresses, same dst MAC address, different src and dst IP addresses (but exluding LAN, localhost, and Multicast addresses, different src and dst port
trPR_v2_100000	100.000	Different src MAC addresses, same dst MAC address, different src and dst IP addresses (but exluding LAN, localhost, and Multicast addresses, different src and dst port

See our published Demo paper presented at IEEE NFV-SDN Conference held in San Francisco 18th-20th, November, 2015.    [Bibtex]

Check our accompanying poster!

Citation: Google scholar