Service Providers

pmoyer

NetIron sFlow Enhancements

by pmoyer on ‎06-18-2013 06:10 AM - last edited on ‎10-28-2013 10:51 PM by bcm1 (331 Views)

sFlow is a very interesting technology that often gets overlooked in terms of network management, operations and performance. That’s a shame; as it can be a very powerful tool in the network operator’s tool-kit. In this brief blog, I hope to shed some light on the appealing advantages of sFlow. To start with - if you are a network operator and you are not gathering network statistics from sFlow, I hope you will carefully read this blog!

 

While the title of this blog says enhancements to sFlow, I’d like to focus a good portion of this piece on sFlow itself and explain why it’s not just useful, but why it should be considered a necessary component of any overall network architecture. I’ll also point out some differences between sFlow, NetFlow and IPFIX (since I frequently get asked about these when I talk about sFlow with customers).

 

sFlow Overview

 

sFlow was originally developed by InMon and has been published in Informational RFC 3176. In a nutshell, sFlow is the leading, multi-vendor, standard for monitoring high-speed switched and routed networks. Additional information can be found at sFlow.org.

 

sFlow relies on sampling; which enables it to scale to the highest speed interfaces, such as 100GbE. It provides very powerful statistics and this data can be aggregated into very edifying graphs. Here is a pretty cool animation describing sFlow in operation. sFlow provides enhanced network visibility & traffic analysis; can contribute relevant data to an overall network security solution; and can be used for SLA verification, accounting and billing purposes. sFlow has been implemented in network switches & routers for many years and is now often implemented in end hosts.

 

Here are some publicly available sFlow generated graphs from AMS-IX (the sFlow samples are taken from Brocade NetIron routers).

 

Here is another simple output from sFlow, showing the top talkers in a specific IP subnet.

 

5.jpg

 

Short Comparison of sFlow, Netflow and IPFIX

 

While sFlow was explicitly invented as an open standards based protocol for network monitoring, Netflow was originally developed to accelerate IP routing functionality in Cisco routers (it remains proprietary to Cisco). The technology was subsequently modified to support network monitoring functions instead of providing accelerated IP routing; however, it can exhibit performance problems on high-speed interfaces. Furthermore, sFlow can provide visibility and network statistics from L2 – L7 of the network stack, while Netflow is predominantly used for L3 – L4 (there is now limited L2 support in Netflow but there is still no MPLS support).

 

Another key difference between the two protocols is that sFlow is a packet sampling technology; while Netflow attempts to capture entire flows. Attempting to capture an entire flow often leads to performance problems on high-speed interfaces, which are interfaces of 10GbE and beyond.

 

IPFIX is an IETF standards based protocol for extracting IP flow information from routers. It was derived from Netflow (specifically, Version 9 of Netflow). IPFIX is standardized in RFC 5101, 5102, and 5103. As its name correctly implies, IPFIX remains specific to L3 of the network stack. It is not as widely implemented in networking gear as sFlow is.

 

sFlow and OpenFlow?

 

There is some recent activity around integrating sFlow with OpenFlow to provide some unique “performance aware” SDN applications. For example, take a look at this diagram:

 

6.jpg

 

[Diagram referenced from here.]

 

In this example, sFlow is used to provide the real-time network performance characteristics to the SDN application running on top of an OpenFlow controller, and OpenFlow is used to re-program the forwarding paths to more efficiently utilize the available infrastructure. Pretty slick, huh? This example uses sFlow-RT, a real-time analytics engine, in place of a normal sFlow collector.

 

NetIron sFlow Implementation Enhancements

 

Brocade devices have been implementing sFlow in hardware for many years. This hardware based implementation provides key advantages in terms of performance. The sampling rate is configurable and sFlow provides packet header information for ingress and egress interfaces. sFlow can provide visibility in the default VRF and non-default VRFs. NetIron devices support sFlow v5, which replaces the version outlined in RFC 3176.

 

In addition to the standard rate-based sampling capability, NetIron devices are capable of using an IPv4 or IPv6 ACL to select which traffic is to be sampled and sent to the sFlow collector. This capability provides more of a flow-based sampling option, rather than just sampling packets based on a specified rate. In addition to sampling L2 and L3 information, sFlow can be configured to sample VPN endpoint interfaces to provide MPLS visibility. Neither Netflow nor IPFIX can provide this type of visibility.

 

One of the new enhancements to the NetIron sFlow implementation is the ability to provide Null0 interface sampling. Service providers often use the Null0 interface to drop packets during Denial of Service (DoS) attacks. sFlow can now be configured to sample those dropped packets to provide visibility into the DoS attack. This feature is in NetIron Software Release 5.5.

 

The other new enhancement that I’d like to mention is the ability to now capture the MPLS tunnel name/ID when sampling on ingress interfaces. This feature is coming very soon and will provide additional visibility into MPLS-based networks.

 

In summary, I hope you gained some additional insight into the advantages of leveraging the network visibility that sFlow provides. One last thing I’d like to correlate to sFlow is Network Analytics. These are complementary technologies which can co-exist together in the same network, while performing different functions. Brocade continues to innovate in both of these areas and I welcome any questions or comments you may have on sFlow or Network Analytics.