Ericsson Evolved IP Network Solution 2015 Update

Ericsson

Evolved IP Network Solution 2015 Update

Multi-Service, LTE-A Readiness and Converged Transport Report

IntroductionFollowing the success of the 2014 EANTC reportfor Ericsson’s Evolved IP Network, the vendorcontracted EANTC to create an updated version for2015 covering additional solution capabilities andnewly available products.

The 2014 report accompanied the initial release ofEricsson’s Evolved IP Network solution. Ericssonexplained that they take a holistic approach indesigning and supporting their end-to-end solution.This approach incorporates protocol and softwareuniformity across any node in the solution.

Ericsson spent the year since our last reportexpanding its IP portfolio, developing new capabili-ties and consolidating the end-to-end solution story.Since the solution has two major software releasesper year, we had the opportunity to explore capa-bilities planned to be released around the sametime as this report. By executing this second testphase, EANTC has now tested the complete Eric-sson Evolved IP Network solution including allelements of the SSR 8000 router family.

In this report we explore the following main points:

• Network design consistent with previous solutiontest report

• True multi-service router capabilities with BNG,IPsec, Ethernet and IP/MPLS services are verified

• Service support for fixed and enterprise customers

• Readiness for LTE-Advanced

• 100GbE interfaces and the new Ericsson SSR8004 router are tested

• Updated Microwave solution MINI-LINK TN istested

• Network Management System (NMS) is extensivelyexplored

Background

Our previous test report highlighted the end-to-endcapabilities of the solution starting with the multi-standard base stations, progressing through themicrowave radios and exploring the IP/MPLS capa-bilities. We also spent time exploring the packetclock synchronization and high availability capabil-ities of the solution.

The 2014 report was not the first time EANTC engi-neers have tested Ericsson's products. Ericsson hasbeen supporting EANTC's interoperability show-cases since our 2008 Mobile Backhaul test event,and over the years has shown interoperability inareas such as clock synchronization, IPv6 IP/MPLSand MPLS-TP. All the reports are available onEANTC's web site.

Test Highlights

Verified multi-service support on Ericsson SSR 8000 family

Measured 6,000 VPLS instances with 996,000 MAC addresses

Demonstrated 96,000 subscribers supported on a single line card

Tested complete service life cycle toolkit capabilities

Recorded phase stability consis-tently under 1.1 microseconds

Confirmed up to 25%a improve-ment in microwave link utilization

a. For 82 bytes frames

EANTC Test Report: Ericsson Evolved IP Network Solution 2015 Update – Page 1 of 12

Having gotten to know the solution well, we pushedEricsson to demonstrate that the Evolved IP Networksolution is truly multi-service by supporting residen-tial, business and mobile subscribers on a singledevice. We also felt that it was imperative toexplore additional clock synchronization functionssince, with the growing deployment of Long TermEvolution (LTE) and early deployments of EvolvedMultimedia Broadcast Multicast Service (EMBMS),a robust packet clock network is essential fortrouble-free mobile network operations.

Ericsson re-stated IP networks as one of theirtargeted areas in their Q4 2014 report, with salesgrowth of over 10% in all targeted areas. Ericssonalso reported twelve new SSR 8000 contractsduring Q4 2014, of which two were for fixednetworks, bringing their total for such contracts wonto 146. It is evident from the breadth of newfeatures and capabilities that Ericsson brought tothis update test, that Ericsson was not simply restingon their laurels. We see the continuous vendorcommitment and it seems that the service providercustomers are indeed responding.

Tested Devices

The network design, including its layering andguiding principles, remained consistent between the2014 and 2015 tests. This not only made our lifesimple when designing the tests, but also attested toEricsson's commitment to their service providercustomers. Clearly, a service provider thatpurchases the Evolved IP Network solution from Eric-sson, would not expect to have to fundamentallychange its network a year later.

What did change though was the range of devicesprovided for test. While in 2014 we had variants ofthe all-outdoor MINI-LINK PT family in the test, thistime around, Ericsson chose to bring the chassis-based MINI-LINK TN in addition to the MINI-LINKPT. The rest of the devices in the access were allEricsson SP 415 or SP 420 routers (the differencebetween the two being port count).

Figure 1: End-to-end Network Topology

MINI-LINK PT

SP 420MINI-LINK TN

Business

RBS

Microwave Link10 Gigabit Ethernet

100 Gigabit Ethernet

Ixia Traffic Generator

Access Network

GPS Antenna

RBS

Management Network

SP 415

Southbound Interfaces

Aggregation NetworkCore Network Management Network

Clock link — Freq.1PPS Link

Anue Freq./Phase Analyzer

Clock Link — ToD

Emulated Access

1 Gigabit Ethernet

SP 420

Access

Access LANTIME M3000

12:50:00

Internet

SP 420

(BNG & IPsec)

IPsecBusiness

DSL

TimeProvider 2700

12:50:00

AggregationAccess

SSR 8010

Emulated Internet

RBS 6501

Grandmaster12:50:00

Microsemi

Meinberg

IPT NMS

Business

SSR 8010

SSR 8010

SSR 8010

DSL

SSR 8004

Boundary Clock

Business

SP 420

SSR 8004


Apart from the notable addition of the newestcommercially available router – the SSR 8004, therest of the devices were the same as in the previoustest. Under the hood however, there were quite afew new components. For the first time we testedEricsson's 100GigabitEthernet line card, the SSCand BNG cards, and an updated alarm/switchfabric card that enables the SSR products to supportphase synchronization on existing line cards.

Test Equipment

This time we executed the complete project atEANTC's lab in Berlin, Germany. Our lab adminis-trator was overjoyed to welcome boxes and boxesof Ericsson devices which were eventually installedin 4 racks. Ericsson sent a team of engineers toBerlin who spent 6 weeks together with EANTC'stest engineers.

The testing was supported by Ixia who provided amultitude of test devices. In addition to the layer 2/3 testers, we also required IPsec and BNG trafficgenerators. Consequently, the Ixia XG12 we hadwas filled with cards such as the Lava AP40/100G,FlexAP 10G, LSMXMDC and STX4. To measure theclock quality we used Ixia's Anue 3500. In fact, alltests in the project, regardless of their network layeror interface speed, were supported by the same testvendor.

The tests required that test engineers from bothEANTC and Ericsson possessed more than singlenetworking layer expertise; it required them tounderstand the interaction between multiple layers.While IP/MPLS traffic was being emulated end-to-end, we also registered BNG subscribers and IPsectunnels – all tasks that would traditionally be takenby different operational teams.

Multi-Service Capabilities

The Ericsson Smart Services Routers (SSR) are devel-oped fully in-house and are built upon the combina-tion of Ericsson's fixed/mobile telecom heritageand Silicon Valley-based IP competency. After wecompleted the test project in 2014, we asked Eric-sson a seemingly innocent question – what is so'smart' about these routers? Ericsson explained thatan example of the smart capabilities is the ability tocollapse different services into a single router (withan appropriate combination of cards).

We were intrigued and, for the purposes of theseupdated tests, asked Ericsson to include these linecards in a typical combination. Ericsson chose toinclude two functions in addition to the IP/MPLSservices that were already planned for the test:IPsec Security Gateway and Broadband NetworkGateway (BNG).

The BNG functionality allows a service provider toterminate residential wireline subscribers on theEricsson router in the aggregation network. In thesame router, IPsec tunnels, expected to be estab-lished by the mobile base stations, could also beterminated. Of course additional functions andapplications could have been added, but we had tolimit the efforts somewhere and felt that the threeservices we could emulate were sufficient to makethe point that the routers are indeed smart.

IPsec Tunnel Scalability

Based on Ericsson’s Long Term Evolution (LTE)network design recommendation, IPsec tunnelsshould be setup between base stations (eNodeB)and the core mobile network. In other words,connectivity over the backhaul should be encrypted.

In the topology used for the test, the aggregationnetwork included Ericsson’s SSR 8010 routerequipped with a single IPsec termination card(called SSC-1). This will be the logical place in thenetwork to terminate both base station connectivity(before handing that traffic to the Evolved PacketCore) as well as residential and business services.

We used Ixia’s PerfectStorm ONE to measure thenumber of IPsec tunnels that a single SSC-1 linecard could terminate and then generated traffic forall tunnels. We established 8,000 tunnels eachusing IKEv2 for key exchange and AES128 fortunnel encryption. In addition, we also verified thatEricsson RBS could establish a tunnel to the sameSSR 8010.

Figure 2: IPsec Performance


Once tunnels were set, we generated traffic in eachof the tunnels and verified that the router is not onlyadapt at terminating IPsec tunnels, but also is ableto forward traffic. We measured 8.16 Gbit/s down-stream and 6.79 Gbit/s upstream of traffic using afixed frame size of 1368 bytes for payload. Wemeasured 3.77 Gbit/s downstream and 2.32Gbit/s upstream using a mix of packet size. TheIPsec test was run using Ericsson’s SSR 14B GeneralAvailability (GA) code due to performance improve-ments for IPsec.

BNG Session Activation Rate

Next we focused our attention on the BroadbandNetwork Gateway (BNG) function residing in thesame Ericsson SSR 8010. The first aspect tomeasure is the rate with which subscribers could beactivated. Said rate is an indication of what theservice provider could expect should all subscribersattempt to start their DSL-modems at the same timeat home. This is a rare event, however, when theevent occurs, after a major power failure forexample, it is a good idea to keep customers happyby allowing as many as possible to register at thesame time.

As the figure below illustrates, we measured an acti-vation rate of 300 sessions per second. The perfor-mance was consistent for both single stack PPPoEsubscribers (i.e. IPv4 only) and dual stacksubscribers. In the latter case, the BNG assignedboth IPv4 and IPv6 addresses to the subscriber.

BNG Subscriber Scalability

Once we measured the activation rate, we used thevalue to also measure the number of subscribers asingle BNG card could support and then generatedtraffic for each of the subscribers. Here we enter-tained three subscriber types:

• Single Stack – Customer with only IPv4 addresses.

• Dual Stack – Customers with IPv4/IPv6 addresses.

• Quality of Service – Customers that use IPv4addresses as well as three classes of service

We were able to reach the advertised number ofSingle Stack subscribers that Ericsson could support– 96,000. Once all subscribers were activate wemeasured throughput of 36 Gbit/s.

The number of Dual Stack subscribers we were ableto reach was 48,000 and the test showed that therouter was then able to demonstrate throughput of24 Gbit/s. The third condition showed the sameresults.

VPLS Services Scalability

Having investigated two services that traditionallyfocus on mobile and residential subscribers, weturned our attention to a business-oriented service:multipoint-to-multipoint layer 2 connectivity. This isthe kind of service that could be realized using theVirtual Private LAN Services (VPLS) we tested.

A router's MAC address and Virtual Switch Instance(VSI) capacities will affect the profitability of theoperator's service offerings. This is because thenumber of services a carrier could sell to customersinterested in multipoint-to-multipoint Ethernetservices is limited by these two factors. The latter istypically limited by the number of end points partici-pating in the virtual switch domain – more endpoints translates to higher signaling load on therouter.Figure 3: BNG dual stack session activation rate

Figure 4: BNG throughput performance


SSR 8010

We designed the test in such a way that 100protected services were initiated by the accessdevices (both SP 420) and were joined by addi-tional sites emulated by the Ixia tester on each SSR8004 in the aggregation network. We asked Eric-sson to configure a total of 6,000 VPLS instances onfour different routers and then used two Ixia ports toemulate additional PE routers. This meant that eachvirtual switch domain included four routers and atotal of 12,000 attachment circuits.

Once the configuration was created (a script-inten-sive activity) we went through the steps ofmeasuring how quickly the routers learn MACaddresses and how many MAC addresses could belearned in total, while also monitoring the CPU loadand memory usage on all devices under test.

The result was a MAC learning rate of 30,000addresses per second for a total of 996,000 MACaddresses. This means that for example all ofOrange Business Services' employees could acti-vate their devices during the same second andimmediately start working. An impressive feat for adevice that's designed as a router and is serving asa switch.

Along the way we discovered a small issue. It wasnot possible to clear the MAC address tables fromthe command line. Ericsson explained that thiscommand will be supported in the General Avail-

ability (GA) code, but as we had an early version ofthe software it was not supported in that build.

We also observed that while learning took place at30,000 MAC addresses per second, a few (under10) MAC addresses took a few more attempts to beinserted into the MAC table. Looking for 10 MACsout of almost a million is a tedious task which is whywe took note, and moved on.

Network and Performance Management

A network service life cycle comprises requirementssuch as service discovery, creation, modificationand trouble shooting. We reported previously thatEricsson was in the process of replacing their NMS.For this year's updated test, we had the opportunityto verify the functionality of Ericsson's IP TransportNetwork Management System (IPT NMS).

As we were testing service lifecycle management,we allowed Ericsson to choose the service typeunder test. Ericsson chose Layer-3 VPN, andexplained this was their primary focus in the currentNMS. Other services could be managed in parts ofthe network, but due to the combination of NMSand router software releases being used the L3VPNwas selected. As the NMS release to support SSR14B was not yet available, Ericsson needed to usethe SSR 14A release.

Service Creation

The first step in the test was to create a L3VPNservice template, and then use this to configure theservice (including BGP on PE-CE links). The servicespan involved two Ericsson devices: the SP 420 andSSR 8010. It took less than 5 minutes to create andconfigure the service. As soon as we activated theservice via the NMS Graphical User Interface (GUI)the service end points and the access circuits weresuccessfully provisioned (roughly within 40seconds).

BNG SSC ENET

SSR 8010

Aggregation Core

6,000 VPLS12,000 Attachment circuits 1Million MAC address

8,000 IPsec tunnels

96,000 Single stack OR48,000 Dual stack OR 48,000 QoS enabled

single stack PPPoE routes

Route Reflector32,000 VPN routes 1,800 BGP label unicast

8,000 L3VPN

Clock synchronizationFrequency offset: up to 3.8 ppbMaximum Time Error: 155 ns

Access

Figure 5: Multi-Service Support in Ericsson EIN

ENET

SSR 8010SSR 8004

ENET

Multi-Service Capabilities Tests Highlights

Support 96,000 Single Stack Subscribers

Support 48,000 Dual Stack Subscribers

Support 8,000 IPsec tunnels

Measured up to a million MAC addresses

Verified up to 6,000 VPLS instances


Once a service is created, it may be necessary tomake changes to it. We extended the same servicethat originally had two end points by adding athird. This is analogous to a customer asking for anadditional site to be connected to an existing VPN.Adding another site required the creation of a newservice based upon the pre-existing service. Thisseemed like a strange approach to a service modifi-cation, especially since once we removed the addi-tional site from the service, the system accepted theremove request, but reported that the service waspartially deployed. This, as explained by Ericsson,was because of the difference on the IPT NMScompared to the active version of that servicerunning on the routers. Once the configuration waspushed to the routers we observed the expectedbehaviour.

The next activity in our service life cycle test wasservice discovery. We already knew that the servicewe created was installed having sent tester traffic toconfirm that everything behaved as expected. Weasked Ericsson to tell us if other services wereinstalled in the network. Ericsson created a servicereconciliation task and let it loose on the network.The task quickly returned with the correct informa-tion: 65 L3VPNs were installed in the network.

Last but not least, we wanted to check the health ofthe service we created. Ericsson's IPT NMS wasconfigured to monitor all service end-points usingPing while we configured the tester to send trafficbetween all end points. The tool reported theaverage latency for the service and showed all end-points were active. We then failed one of the accessdevices (emulated by Ixia) and were immediatelynotified by IPT NMS of the issue. The alarm wasspecific enough that we knew not only that therewas a problem with the service, but also exactly

where the problem was (to a slot/port level). Wecould tell that this was immensely useful yet weasked ourselves what would happen when a serviceof 1,000 end points had to be monitored. If 1,000ping pairs would have to be created, then clearlythe usability of the tool would be limited, unlessanother approach is used.

Network Performance

In the 2014 report we tested two of Ericsson'sEvolved IP Network products at full scale – the SP420 and the SSR 8010. For this year's test weturned our attention to Ericsson's newest product –the SSR 8004.

Alongside the new router, we also expanded ourinvestigation into one of Ericsson's key productareas: Microwave transport. Ericsson has deliveredover 3 million MINI-LINKs1, and has a market shareof 25%, which tells us they take their microwaveproducts very seriously! In this segment of the testswe really looked into throughput as well as optimi-zation methods.

Figure 6: L3VPN Service Creation Using IPT NMS

1. http://www.ericsson.com/news/140320-mircowave-milestone_244099438_c

Figure 7: L3VPN Service Monitoring IPT NMS

Network and Performance Management Test Highlights

IPT NMS covers the complete service life cycle


http://www.ericsson.com/news/140320-mircowave-milestone_244099438_c

http://www.ericsson.com/news/140320-mircowave-milestone_244099438_c

SSR 8004 100GigabitEthernet Performance

For the throughput performance test focusing on theSSR 8004 we asked Ericsson to mix two line cardswhich we believed to be representative of therouter’s position in the network: 10GigabitEthernetand 100GigabitEthernet. The logic used for the testwas that today the 10GbE interfaces will face theaccess network, while the 100GbE interfaces willface the core network.

The tests followed the traditional IETF-defined RFC2544 for throughput measurements. Based on yearsof experience we ran the tests twice – once tomeasure the throughput using 30 second tests andthen, using the values collected in the short test run,repeated the tests at each frame size for 10minutes. We expected the results, based on thenumber of ports in the test, to be 400Gbit/sthroughput with latency in the order of tens of micro-seconds.

Before the tests started Ericsson explained that ourexpectations may not be accurate. As a conse-quence of their design choices on these line cardsEricsson, does not consider wirespeed forwardingof smallest packet sizes as a realistic real world usecase.

Our tests confirmed Ericsson’s statements. We wereable to measure line rate performance at packetsizes of 373 and 1,518 bytes (with latencies of 43and 51 microseconds respectively), but for thesmaller packet sizes of 70 and 128 bytes, wemeasured 53 and 81 percent of theoretical linerate.

We also ran one test with a mix of packet sizes(called IMIX). Here the tester is configured to send aseries of packet sizes with varying weights. Sinceour IMIX was heavily weighted towards smallpacket sizes (72% of the packets were smaller than256 bytes) we were only slightly surprised to seethat we could reach 96.87% of line rate. We alsoobserved that latency increased in this test case byup to 1,046 microseconds. Ericsson explained thatthe increase in latency was due to larger packetsdelaying the transmission of smaller packets placedbehind them in the queues.

MINI-LINK Throughput

One way to think about the MINI-LINK microwavetransport solution is as an Ethernet switch, wheresome of the interfaces are not implemented withcopper or fiber. Instead they are based on a radiointerface that brings new challenges switches do notnormally face, yet must still perform as any otherswitch would be expected to in a transport network.To measure the performance of the switch part ofthe MINI-LINK we ignored the microwave interfacesand connected all 8 GigabitEthernet ports to thetester and ran the same standard RFC-basedthroughput test we ran with the Ericsson SSR 8004.

We used the same packet sizes and the same dura-tion and received full line rate results for all fixedframe size tests of 30 seconds duration. Whenrepeating the tests with IMIX traffic, we measured66.65% of total line rate or 5.25 Gbit/sthroughput. In this case we also measured increasein latency from 515 ms for 1518 byte packets to996 ms.

We repeated the test for 10 minutes to verify thatthe switch will be stable over longer operationalperiods. In this condition, for each of the framesizes (fixed and IMIX), we recorded minimal packetloss of 0.001%.

Figure 8: Ericsson SSR 8004 in the test


Microwave Deep Buffers

The majority of the traffic on the internet today runsover TCP, which utilizes flow control and congestioncontrol mechanisms in order to provide reliable end-to-end transport. TCP is a demanding protocol - ittries to maximize the amount of throughput itreceives. Should TCP detect packet loss, it backs upby 50% and slowly tries again.

Microwave links in networks therefore run the risk ofbecoming a bottleneck, as even if a network hasbeen designed to take into account available linkbandwidth, congestion can occur due to adaptivemodulation changes (for example caused by heavyrain). To cope with this, Microwave devices shouldbe capable of buffering sufficient frames to be ableto continue sending until the sender receives a TCPacknowledgement.

Ericsson explained that the MINI-LINK PT productshave up to 8 megabytes of buffer to maximizemicrowave link utilization. In this test we measuredthe benefits of these buffers on end-to-end TCPthroughput. As HTTP traffic was used for this test,we used the correct 'goodput' term to indicateapplication level throughput.

In the test setup we created a microwave hop usingtwo MINI-LINK PT, then connected an Ixia port toeach device. We configured a large HTTP object of2 gigabytes and asked Ericsson to use the highestmicrowave modulation of 1024 QAM. We expectthe link, with its 100 ms round trip time, to provideus with 435Mbit/s of goodput.

Initially we configured the microwaves with a tradi-tional small buffer size of 168 Kb. The goodput wewere able to measure was 308 Mbit/s. Once westarted increasing the buffer to 1 Mb and then 1.4Mb, we measured 388 and 435 Mbit/s respec-tively. Increasing the buffer beyond 1.4 Mb led tono further improvement in goodput for our testsetup, as at this point the buffer depth was sufficientfor the round trip time.

Reducing the amount of memory in order to attemptto push down the price of a device, as well aslatency, may be common practice, but not for Eric-sson. As everything in life, there are tradeoffs to bemade. The trade-off in the case of reducing latencyand memory can lead to a false economy whenconsidering LTE and LTE-Advanced deployments.These require not only ever-increasing bandwidth,

but also ever-increasing goodput. It is thereforeimportant to consider the end-to-end transport solu-tion including microwave, in order to deliver thequality of experience subscribers are looking for.

Microwave multi-layer header compression

Another method to squeeze more bandwidth from anetwork link is to compress the traffic itself. Ericssonhas implemented multi-layer header compression intheir microwave products and asked us to test theMINI-LINK PT 2020 to verify that compression reallyprovides meaningful performance improvements.The compression method used here is multi-layersince it compresses headers such as Ethernet, MPLSand IP. Ericsson claims that this compression willhave the greatest impact in mobile networks sincethe payload itself is likely to be compressed orencrypted by other elements in the network.

From a testing perspective, the use case was simple.We generated bidirectional traffic across a micro-wave link and ran the test with and without thecompression feature enabled. We also used twosets of traffic conditions, an internet mix of packetsizes ranging from 64 bytes to 1518 bytes (with72% of the packets being smaller than 129 Bytes)and the second with only 82 bytes where weexpected to see a bigger impact due to the ratio ofheader to payload size.

We expected to see more throughput in both cases,but just how much was unclear. While throughputwithout compression showed 457 Mbit/s across themicrowave link, throughput with the Internet mixtraffic improved by only 1% to 462 Mbit/s. Theimprovement was however huge for the smallpacket size – 18% to 542 Mbit/s. The reason for

Figure 9: Goodput Results with 1.4 Mb Microwave Buffer


the difference was obvious. An MPLS packet size of82 bytes includes 4 bytes of MPLS headers as wellas 20 bytes of IP headers and 8 bytes of UDPheader. This means that 39% of the packet size isheaders. As we demonstrated, these headers couldbe compressed and since such a large portion ofthe packet is headers, the gain of compression ispronounced.

Compression for mobile traffic could take place intwo locations in the network – in the access andafter the Evolved Packet Core (on the SGi-interface).Ericsson argues that application acceleration shouldbe used on the SGi-interface while Microwavecompression, agnostic to the traffic payload, is thebest way to achieve efficiency in a mobile accessnetwork.

Network Availability

It seems that every test we ever execute includeshigh availability test cases. This is an indication ofhow important networks have become to our lives,hence operators' sensitivity to high availabilityperformance. These days, with so many criticalbusinesses operating on the Internet, we simply cannot afford to lose connectivity and hence want thenetwork to be as robust as possible.

We had already successfully tested IP Fast Rerouteon the SSR routers, Bidirectional Fault Detection(BFD) between the SSR and SP routers, as well asgraceful restart on the SSR in the 2014 activity -therefore we chose to focus this time on new capa-bilities of the Evolved IP Network solution.

IP Fast Reroute: Loop Free Alternates on SSR

Despite all of the success that MPLS has seen in thelast 15 years, some argue that the protocol stackhas become too bloated. To operate an MPLSnetwork with all its bells and whistles, one mustdeploy RSVP-TE, an IGP, BGP, and label distributionprotocol (LDP). In recent years the IETF has beenstudying if there is a way to reduce this list withoutlosing capabilities.

IP Fast Reroute is part of the answer to this question.The goal of the RFCs that define IP Fast Reroute is to“reduce failure reaction time to 10s of millisec-onds...” (RFC 5286), and achieve this without theuse of RSVP-TE by simply pre-computing an alter-nate next-hop that is activated when the primarynext-hop fails.

Ericsson explained that at the time of the testing thefeature was only available on the SSRs which iswhy we focused our attention on the aggregationnetwork and emulated the failure between SSR8004 and 8010. We ensured that no RSVP configu-ration was in use on these routers while weexecuted the test, only MPLS with LDP fed from IS-IS.

We emulated two failure scenarios in the test:optical layer failure in which BFD was used todetect the issue and the traditional loss of signalfailure. Each failure scenario ran three times testingboth failure and recovery.

In the test runs which used BFD to detect the failurewe measured 107 to 111 ms out of service timebetween test runs. This certainly was an impres-sively consistent result. In all recovery tests we alsorecorded some out of service times in the range of19 to 27 ms.

In the tests that used loss of signal to detect failurewe measured 6 to 24 ms of service disruption.Again, during the recovery phase we measured upto 6 ms service impact.

Ericsson explained that the BFD times should havebeen much lower (as seen in the previous report),however, a bug in the pre-release software that wewere using meant that BFD-detected failures werenot triggered as quickly as the loss-of-signal usecase. This bug was known before the test was run,and Ericsson confirmed it is fixed in the GA soft-ware version.

LFA tool

Together with the other service life-cycle tools wediscussed, Ericsson demonstrated an off-line IP FastReroute (FRR) Loop-free Alternate (LFA) analyzer,which is integrated with Ericsson's NetOp EMS.Using this GUI, we created a ring network topologywith four nodes, LFA converge metric, non-protectedroutes as well as links for the destination nodes. Thetool allowed us to look into “what if” scenarios,identify where in the topology potential resiliency

Network Performance Tests Highlights

100GigabitEthernet line rate for 373 and 1518 bytes packet size

Up to 25% improvement in Microwave link utilization through compression


bottlenecks would occur, as well as providesuggested optimized topologies.

RSVP on SP 400

In the previous report we tested the failure in theaggregation and core networks. This time wefocused our attention on the access network. SinceEricsson added support for Resource ReservationProtocol - Traffic Engineering (RSVP-TE) in its SP 400devices they could now protect MPLS tunnels origi-nating in the cell site routers. We were more thanhappy to oblige in testing this functionality.

Ericsson chose to configure static backup tunnelswith explicit routes which, if one is looking toadhere strictly to our industry's nomenclature,means that the test was not verifying RSVP-TE FastReroute. Still, the test verified the solution's ability toprotect MPLS tunnels in the access network.

We chose a link between the Ericsson SP 420 andthe SSR 8004 and disconnected it while traffic wasflowing for one of our services. We then measuredthe out of service time based on the number of

packets that were missing and verified that theservice indeed returned to normal operation. Werepeated the test three times in order to rule outoutliers.

The results satisfied Ericsson’s expectations. In theupstream direction the highest measured out ofservice time was 256 ms with the lowest being 248ms. In the downstream direction we measuredbetween 127 and 84 ms out of service time.

We also evaluated if a recovery from the failurewould result in service disruption. We reconnectedthe link between the Ericsson SP 420 and the SSR8004. No packet loss was observed at all in thedownstream direction. In the upstream direction,leaving the SP 420 towards the SSR 8004, amaximum out of service time of 89 ms wasrecorded. Ericsson explained that the minimal lossof traffic, which was expected (as per the test plan),was the result of hardware differences between theSP 420 and the SSR 8004.

LTE-A Readiness

In our 2014 test we spent a great deal of reportingreal estate on clock synchronization. We demon-strated that the Ericsson Evolved IP Network solutionwas ready to support Long Term Evolution (LTE)deployment and included not only clock accuracy,but also clock robustness.

That was 2014. The conversation amongst mobileoperators has moved on to the higher capacitiesand lower latency that both LTE-Advanced and 5Gare promising. For these the importance of clockaccuracy, stability, and robustness become evermore pronounced. Ericsson was eager to demon-strate that their Evolved IP Network solution is morethan capable of meeting such requirements.

There is a plethora of new aspects in these tests.Since the tests were performed at EANTC's lab,where, at the time, both Microsemi and Meinberghad GrandMaster clocks installed, Ericsson wasable to demonstrate a dual-grandmaster clockstrategy. Each test performed in this section was

Figure 10: Network topology before optimization

Figure 11: Optimized topology for link protection

Network Availability Tests Highlights

Loop Free Alternates in Ericsson SSR supported

RSVP convergence of up to 256 ms in SP 420


executed twice, once using the Meinberg LANTIMEM3000 as clock source and again with theMicrosemi 2700 GrandMaster.

Ericsson explained that to enable IEEE 1588 on theSSR requires only an updated version of the ALSW(Alarm and Switch Fabric) card. The card can beupgraded in-service without disruption to anexisting network. All SSRs required to distributeclocking information were equipped with this newcard (ALSW-T), so what did we actually test?

Full Path End-to-End Clock Synchronization

Full path timing support is defined as each node inthe network playing a role in the clock deliveryservice. In Ericsson's case that meant all routers andmicrowave devices between the GrandMasterclocks and the RBS were configured to function as aBoundary Clock.

In line with ITU-T G.8273.2, Ericsson configuredtheir devices to use physical layer frequencysupport from Synchronous Ethernet, with phaseinformation distributed using the Precision TimeProtocol (PTP) profile specified by G.8275.1. Thegoals for the test were in line with the standards andfrequency deviation of no more than 16 parts perbillion (ppb) with phase error within ±1.1 s. Thestandards actually call for phase error limits within±1.5 s, but Ericsson explained that the additional400 nanoseconds is an air interface budget.

We ran the test without any traffic in the network for15 minutes once the slave clock showed a lockstate. The results of this test, which we consideredas baseline, were great. The frequency offsetranged between 1.6 and 3.8 ppb. The maximumtime error ranged between 124 and 155 nanosec-onds – an order of magnitude better than expected.

Long Term Clock Stability

After the baseline test was completed we movedtowards a more realistic test case, as no network isexpected to run without traffic. This time we usedthe industry standard measurement guidelines fromITU-T G.8261 to generate traffic in the network. Weused the two most often referenced test cases, 12and 13, and ran the tests for more than 1 hour andmore than 14 hours respectively.

We measured frequency offset of at most 6.56 ppband maximum time error of up to 157 nanosec-onds. The change between a 'clean' network and anetwork with emulated traffic was therefore almostnon-existent, highlighting the benefit of the fulltiming support model.

Clock Robustness

Now that we have seen how the network works innormal conditions, we asked Ericsson to demon-strate what happens when the network is experi-encing failure. In our network there were twoobvious failure scenarios – a failure in the linksbetween the routers and a failure caused by micro-

Figure 12: Packet Clock Synchronization – Full Path Support

GPS

RBS 6501

Antenna

Microwave Link 10 Gigabit Ethernet1 Gigabit Ethernet

1PPS LinkClock ReferencePTP Distribution Path SyncE Distribution Path

Access Aggregation Core

SSR 8010

MINI-LINK TN

Grandmaster12:50:00

Clock Link — Freq.Boundary Clock

SSR 8010

SSR 8010SP 415 SP 420

SP 420 SSR 8004

SSR 8004

SSR 8010

Meinberg LANTIME M300012:50:00

Ixia Anue 3500 Frequency/Phase Analyzer

Microsemi TimeProvider 2700 12:50:00


wave nodes losing their air interface. Bothscenarios require that the last node implements afunction called holdover which allows the node tocontinue delivering a clock signal withoutannouncing to the world “I am lost!” andsuppressing the clock output altogether.

This year we also had a small cell radio basestation from Ericsson, the multi-standard outdoormicro base station known as RBS 6501, whichparticipated in the packet clock distribution infra-structure and was in fact the last node in the chain.While we could not measure the clock quality onthe mobile air interface (for lack of instrumentation)we did verify that the RBS was reporting a lock tothe clock signal during all of the test runs, provingthat it remained locked while the transport networkwas in a holdover state.

The first scenario investigated the microwavefailure. Here we simply disconnected the RF cablesbetween the two antennas while measuring theclock quality. We also verified that the devicesnever reported “Free-running” state.

Shutting off the Microwave link actually meant thatthe SP router, positioned in front of the microwaverouter had to perform the hold over function. In bothtest scenarios (i.e. with both GrandMaster clocks)we were able to measure frequency deviation of atmost 6.12 ppb. The maximum time error recordedwas 191.2 nanoseconds.

We also used this opportunity to investigate whathappens when the microwave was not completelyturned off, but the frequency modulation wasreduced. We used an attentuator to repeat the testtwice: once with 512 QAM and once with 16QAM. Both tests showed that the clock signal wasnot affected by modulation changes and the valuesrecorded were along the lines of all other tests.

The results of the link failure between the SP 420and the SSR 8004, the last failure condition wemeasured, were similar to the above. The maximumtime error we recorded was - 309 nanoseconds.Again, the RBS reported remaining locked, andnone of the devices went into holdover mode.

Ericsson Evolved IP Network Solution Tests Summary

We applaud Ericsson’s commitment to continuoussolution development and independent validation.We are also pleased to see the Evolved IP Networksolution continue to progress and report success inthe market.

With this updated report we can confirm that Eric-sson’s Smart Services Router (SSR) lives up to itsname and is capable of delivering truly convergedand concurrent multi-services such as BNG, IPsec,Ethernet and IP/MPLS.

Multi-service is not the only capability highlightedhere, but is a critical element in Ericsson’s commit-ment to being a dependable solution to their serviceprovider customers as they grow their networks.Looking at the clock synchronization results as wellas the integration of the base station with the trans-port infrastructure, we get a definite sense that Eric-sson is planning to provide their customers with averified solution that anticipates the challengesinvolved with 3GPP and 5G migration.

About EANTC

The European AdvancedNetworking Test Center(EANTC) offers vendor-neutral network test servicesfor manufacturers, serviceproviders and enterprisecustomers. Primary businessareas include interopera-bility, conformance and

performance testing for IP, MPLS, Mobile Backhaul,VoIP, Carrier Ethernet, Triple Play, and IP applications.

EANTC AGSalzufer 14, 10587 Berlin, [email protected], http://www.eantc.com/vF1.1 20150315, JG

LTE-A Readiness Tests Highlights

Phase accuracy consistently under 310 nanoseconds including failover scenarios

Full on-path clock synchronization support

Microwave adaptive modulation does not affect clock performance


Documents

Ericsson Evolved IP Network Solution 2015 Update