Via IP Converged Services Case Study
| October 18, 2010
Category: Case Studies
Abstract
Via IP is specialized in proving innovative managed corporate networks. Via IP excels in connecting LANs and WANs, maintaining bandwidth efficiency, managing data from different protocols and reducing overhead costs for your business.
This case study examines how Via IP provided a highly redundant Voice over IP ready network that seamlessly intergraded with the campus network of one of our key clients, and demonstrates the additional layers of intelligence Via IP overlay onto the network to ensure it meets the client’s requirements.
Network overview
Via IP installed a managed MPLS Virtual Private Network WAN that would connect the clients campus network to the Via IP core, to extend the reach of the clients campus to branch offices connected on the Via IP V+one network in addition to the clients co-located resources in the Via IP data centre and the public Internet. A core requirement of the managed WAN, was that the campus network should be connected to the core by means of fully redundancy links. Via IP accomplished this by providing a redundant pair of WAN links each using disparate technologies into two geographically separate points on their campus.
The first link was a high speed Private Network over Wireless (PNoW) circuit installed in a remote region, supplemented by a lower speed Private Network over Copper (PNoC) service at the headquarters. By using disparate wired and wireless WAN technologies, these two services where combined to provide diverse path connectivity to the primary network. Furthermore using dynamic routing Via IP will aggregate the bandwidth of each link and provide unequal cost load balancing where data is passed over the two links in proportion to the available bandwidth. In practice, roughly twice as much data will be sent over the high speed Private Network over Wireless (PNoW) circuit as is sent over the headquarters link.
In the event of a failure on either link for any reason, the network will reconverge quickly and the remaining link will carry all the traffic until communication is restored on the first link. After a period of stabilising, the network will automatically reconverge and again start to use both links.
In addition Via IP also houses the managed Voice over IP phone system in one of the Sydney Data Centres. The Voice over IP phone system equipment is also connected to the private network which ensures full connectivity without requirement or use of firewalls.
Each of the Private Network links terminates on a Cisco router proactively managed by the Via IP Network Operations Centre.
Converged Services
The private network links have been designed to carry converged data and voice services to allow for VoIP to be carried simultaneously with data traffic. VoIP is a real time application and as such has low tolerances to delays in transmissions. For this to be possible the Via IP links have been carefully sized to allow voice calls to coexist on the same network as the data traffic. In additional low latency QoS techniques have been employed across the Via IP network to allow end to end quality of service to protect voice calls.
The following sections outline the design that allows converged voice and data to be carried over the Via IP network.
The amount of network bandwidth a voice call requires is largely dictated by the codec in use. The codec (coder/decoder) converts the analogue voice stream into a digital signal so the voice call can be carried over the IP network. There are many types of codecs and each has their own bandwidth requirements.
The G.729 Codec has been selected because it only requires in the region of 30 Kbps per call depending on the compression techniques employed. The network links have been sized such that they are able to carry the sum of codec requirements as detailed at the time of design.
To ensure that the voice traffic is transmitted over the Via IP network in a timely fashion, Quality of Service (QoS) is employed to ensure all voice traffic is transmitted with the highest priority. QoS is designed on the Via IP networks end to end, so that every network device is QoS aware and honours the QoS policies.
The most effective way to achieve the desired QoS policy is to have the QoS trust boundaries set as close to the end points as possible. In VoIP deployments this means the VoIP handsets. To provide the end to end QoS policy, the Via IP routers trust the Differentiated Services Code Points (DSCP) Expedite Forwarding (EF) tag the VoIP handsets add on to the VoIP packets as they transmit the voice call over the network.
The Via IP routers check all incoming packets for the DSCP EF tag. Packets that are tagged DSCP EF are then added to routers priority queues where they are guaranteed bandwidth. These packets are sent as a priority over other normal data packets which are not time sensitive. The DSCP tags remain on the voice packets as they traverse the Via IP network so that every Via IP router will prioritise the voice calls.
Voice quality can be measured by an industry standard known as the Mean Opinion Score (MOS). This is a method to report the subjective quality of a voice call from the end users perspective. MOS is expressed as a value between 1 and 5. 5 being the best. A MOS value of 5 would represent a face to face conversation. A MOS score above 4 represents a normal voice call. MOS values between 3.6 and 4 corresponds to acceptable quality levels.
Using QoS and adequately sized links for the maximum number of concurrent voice calls, Via IP endeavour to achieve a MOS score of 4 or above. The MOS score is graphed through the Via IP Halo monitoring platform.
In Voice over IP (VoIP) networks the three most important key performance metrics that impact voice quality are latency, jitter and packet loss. Latency describes the time it takes for one packet to travel from its source to destination over the network. Latency can be recorded as a one way or return trip which measures the time taken to send a packet and for that packet then to return to the sender. The term jitter refers to fluctuations of network transmission delays which can cause irregularities in the voice call. Packet loss is measured in the percentage of packets that are lost during a voice call.
In accordance with industry best practices Via IP aims to achieve packet loss less than 1%, round trip delay of less than 300 ms and jitter less than 30ms.
The voice over IP telephony system was installed at Via IP’s co-location data enter by one of Via IP’s key partners who specialise in large scale corporate multimedia contact and converged IP telephony systems for business enterprises.
A key part of the design was the use of dual high speed links connecting the campus to Via IP’s V+one network. Using dynamic routing protocols the two links will operate as a fault tolerant pair, with each link providing backup functionality for the other. The underlying technology of the two links is different so as to achieve diverse path connectivity.
Voice over IP traffic has specific requirements in terms of delay and jitter. As voice over IP traffic is UDP based, it has no sequencing function, therefore packets must not arrive out of order. To this end Via IP has engineered the network so that voice traffic will only traverse a single link in any one direction of flow, whilst still allowing data to flow over both links in a load balanced fashion therefore utilising both links and maximising the return on investment.
If for any reason the main link was to fail, voice traffic will fail over to the second circuit. Both links are QoS enabled to ensure voice traffic is prioritised over data.
After a failure on the 10 Mbps link has been restored, voice traffic will pass back over the 10 Mbps link only and data traffic will once again be load balanced.
Redistribution
Due to the requirements of the way that voice must be handled over redundant load balanced links a complex routing design is required.
The client uses OSPF their campus to advertise their routes throughout their redundant enterprise. To fulfil the client’s requirements, Via IP use EIGRP between the managed CPEs and core to allow for unequal cost load balancing (OSPF does not support unequal cost load balancing), and OSPF between the managed CPEs and the campus.
As the Via IP managed CPE routers run both OSPF to learn the client’s routes and EIGRP to learn the Via IP prefixes, redistribution is needed between the two routing protocols. Redistribution is the process by which routes learnt by one particular routing protocol can be leaked into another routing protocol.
In this instance the two Via IP managed CPE routers learn Via IP connected routes, which include the data centre hosted Voice over IP telephony equipment and the ADSL connected branch offices via EIGRP. The same pair of managed CPE routers learns the client’s campus routes via OSPF Area 0.
Mutual redistribution is performed allowing OSPF routes to be redistributed into EIGRP and EIGRP routes to be redistributed to OSPF. However mutual redistribution can cause routing inconsistencies, especially where load balancing over the redundant WAN links needs to be preserved.
Mutal redistribution in a fully meshed network can however cause sub-optimum routing, therefore some controls are necessary. Sub-optimum routing can occur when the Administrative Distance (AD) of routes change as they are redistributed. For example an EIGRP learnt prefix has an AD of 90 and OSPF prefixes an AD of 110. However a prefix learnt by EIGRP which was introduced into EIGRP via redistribution has an AD of 170. As that route with an AD of 170 is redistributed into OSPF, the AD changes to OSPF's default of 110, and therefore seems to the network to be the more attractive route irrespective of cost and therefore has an impact on the routing topology. To minimise this affect, several advanced routing protocol techniques have been deployed in this design.
Collocation Services
Via IP provided the client with ½ Rack (20RU) of co-location rack space located in the Global Switch Sydney data centre.
The data centre has full UPS, redundant environmental control, power and fire detection and suppression systems. The rack has dual power bars which in turn have dual UPS and generator protected feeds. This extends power protection to any device that has dual power supplies.
A single 100 Mbps Fast Ethernet port connects the rack directly onto the backbone of the MPLS Virtual Private Network. There is no filtering between this port and the rest of the Private Network. The rack has a power allocation of 1KW averaged over the calendar month.
Conclusion
In this case study Via IP demonstrates how they implemented a fully redundant, self healing intelligent WAN topology, which whilst necessarily complex in design, is simple in its execution requiring no changes on the client’s OPSF enabled campus, to allow for the load balancing of data traffic whilst preserving the integrity of voice flows.
