Product cycles being what they are, we often see a peak in launches and announcements in the autumn conference season, with another round at or before MWC. This week, with Broadband World Forum in Amsterdam to the fore, we’ve seen a clutch of product announcements or launches.
They encompass new technology implementations to boost capacity, to software for planning and optimisation, to advances in vendors’ approaches to NFV and SDN use cases.
Here, then, in no particular order, are the key announcements that crossed my inbox this week.
Ericsson announced it had a line card upgrade path for Smart Services Router, announcing it had a new board for the router based on new processing capabilities supported by its in-house developed network processor – the SNP4000.
Bit of a snooze? Maybe not.
Something like the SSR is required to perform a lot more functions than an edge router of a few years back. The SSR might be required to implement application awareness using DPI, perform LTE gateway functions and terminate IP Sec traffic, network address translation, perhaps firewall and IDS. That’s a lot more than the traditional L3 packet processing.
In turn, that puts a strain on “traditional” network processing units, that are, according to The Linley Group, reaching “breaking point” due to restrictions in code flexibility, memory and storage, and the ability to provide L4-7 packet processing and inspection.
(If you want to read more about this, see this whitepaper from Linley that is hosted on Ericsson’s site – I don’t know if it was paid for by Ericsson)
You’ll see from that paper that, in Linley’s view, What Ericsson is doing is implementing a high performance processing architecture that enables it to combine the power efficiency of an NPU with the processing power of multi-core CPUs with networking accelerators.
Linley Group says: “We estimate the newest multicore processors deliver about 1Gbps/W, whereas the newest NPUs provide nearly 5Gbps/W. In an edge-router line card, this difference translates directly to network-port density.”
According to Linley, Cisco developed the first NPU with its QuantumFlow Processor (QFP) for the ASR 1000, which was introduced in March 2008.
“In 2H12, Cisco shipped a second-generation QFP using 64 CPUs at 1.5GHz. Using a pair of these packet processors plus a pair of traffic managers, the ASR 1000 is now capable of 100Gbps throughputs. This configuration’s maximum packet rate, however, is only 58Mpps, or slightly less than 40Gbps for 64-byte packets.”
Ericsson’s NPU combines 200Gbps of packet processing with a 100Gbps hierarchical traffic manager on a single chip. The SNP 4000 is designed for line-rate processing at these throughputs, meaning it is capable of processing 300Mpps, as Table 1 shows. The first product based on the new chip, which will be available in 4Q13, is the 20-port 10G Ethernet line card for the SSR 8000 [that Ericsson announced this week].
By supporting a GNU tool chain and SMP Linux, these new Ericsson and EZchip NPUs enable flexibility similar to that of multicore processors. This means the chips can handle virtually any new protocol or feature and adapt more quickly to carriers’ requirements.
For example, the NPUs can implement OpenFlow for SDN and adapt to new versions;
handle multiple applications for service chaining; provide stateful processing for features
such as NAT and firewalls; deliver application visibility for billing, SLA enforcement, or
bandwidth management.”
This focus on system architecture seems a bit dry, but vendors and operators need to ask, where do the functions performed by dedicated hardware migrate to?
This focus on system architecture seems a bit dry, but vendors and operators need to ask, where do the functions performed by dedicated hardware migrate to? Is there a middle ground between standard hardware and high end Ethernet switches? If so, what am I going to run those functions on? It appears as if Ericsson thinks that for now the answer to both these questions might be an edge router, built on a high performance (SNP4000) platform, running virtualised functions.
That’s part of its Service Provider SDN and cloud systems messaging, something it is talking about in terms of developing over the next few years. Is it the standard hardware-based, software defined, open network? Up for debate, but with 12 SSR contracts signed up in the last quarter, this sort of distributed processing capability is hitting a few operator buttons right now.
Sticking with matters SDN, Huawei claimed to have launches the first orchestration network management system for SDN controllers – NetMatrix. Details were sketchy, by in amongst the buzzword density, the company was hitting on the important point of how operators will control OSS integration across networks elements. In Huawei’s case, it said NetMatrix would do this by “shielding” complexity in the network – in this case presenting the orchestrator as simplifier and coordinator of different systems.
NSN is in the midst of a series of product and technology announcements. It did the same thing last year, outlining a programme of announcements under a unifying theme. Here’s something I wrote on last year
