Traditionally, applications remained in-house at fixed locations protected with a well defined perimeter. To enhance user experience (customer, employees and remote workforce), enterprises used WAN optimization techniques, for example techniques like TCP optimization, compression and caching. This worked well with relatively small capacity WAN links. All these techniques worked in favour of ICT, as the cost of bandwidth was exceptionally high when compared to the cost of deploying WAN optimizers.
Enterprises with distributed locations across the world deployed these WAN optimizers to increase the efficiency of WAN and deliver better user experience. WAN optimization techniques have become irrelevant in the current scenario as the cost benefit of deploying these are no longer valid. It worked out to be far more inexpensive to increase the bandwidth rather than deploy WAN optimizers to optimize and increase the efficiency of the existing bandwidth. However the scenario today is significantly different.
Enterprise demand a constant supply of bandwidth. Cloud and Internet are the norms today in the world of digital. Any kind of internet congestion means your business takes a massive hit in its bottom line. According to Gartner, the average cost of network downtime is around $5,600 per minute. That is around $300,000 per hour. Apart from the monetary costs, IT downtime can significantly result in
- Overall business productivity
- Transaction/sales losses
- Loss of Data
- Customer dissatisfaction
- Restarting/return to full operation
- Damage to the company’s brand and reputation
- Regulatory compliance effected
Given that the Internet is the only transport that connects your enterprise to end-users (Customers and Employees) irrespective of their location, and enterprises have no control over internet performance and availability, the link load balancing technique is helping enterprise to ensure an always UP network.
What is Link Load Balancing?
Link load balancing is a technique to mitigate WAN issues like availability and performance to deliver last mile internet connectivity. It increases network performance and resilience drastically by leveraging available WAN links (MPLS circuits, Leased Internet lines, Broadband, and wireless) into a single logical interface. The system actively monitors the quality in terms of latency, packet loss and jitter, and the capacity in terms of the throughput, of the links and steers traffic to the most appropriate link based on the business intent defined as a policy. This technique increases the scalability and effective utilization of network resources.
This technique mainly focuses on WAN Aggregation for 24/7 continuous connectivity and service level assurance for corporate and cloud-based applications. This technique can be deployed in entry level to high-end data centers as well as enterprise branch offices. So how link load balancing was done traditionally? Traditionally WAN link load balancing followed a static definition of how the WAN links are utilized. It required configuration which was applied to state what kind of traffic is made to traverse using the WAN link. It did not consider the current state of the WAN link with respect to availability or performance.
For instance, consider two WAN links. Let us say Link A has an expected capacity of 10 Mbps at a location because it is the bandwidth that the enterprise has subscribed to and Link B has an expected capacity of 5 Mbps at the same location. At the current instant, the measured throughput on Link A is 6 Mbps on a theoretical throughput of 10 Mbps while measured throughput on Link B is 4 Mbps on a theoretical throughput of 5 Mbps. With a static load balancer, the typical configuration would have been to use Link A for traffic class A and Link B for traffic class B. However, if you consider the current performance of the links, the Link B is experiencing less congestion and packet drops compared to Link A. Link B should be the chosen link for all traffic classes unless the Link B cannot support a flow based on the available bandwidth. In the case of WAN Link Aggregation (WLAG), every WAN link is normalized using link capacity and quality. The Cloud scale, peer-to-peer probe mechanism provides link capacity and quality data. Every flow is sampled every 50ms to figure out the network requirement. Flows are re-balanced across available links, every time network condition changes. Load balancing works even if multiple links satisfy the network criteria for your application. Packet replication is used for critical applications like Voice, when all links are below the desired quality.
What has worked in the past will not work in the future!
Reliability and high up-time have always been a priority for Enterprises when it comes to Branch connectivity. Almost every medium to large Enterprise to date has utilized MPLS circuits to connect branch offices to the data center or the central office with a fairly stable, service level agreements driven private connectivity. However, this has primarily depended on the quality of the last mile link. The network IT team has always relied on the service provider or multiple service providers to provide for redundant links at the last mile. These redundant links have acted as a fail safe, only to be used in case of primary link failures.
The issue with this approach has been a waste of available bandwidth as one of the links always gets underutilized as it is only present to provide for redundancy while the primary link that gets overly utilized leads to bandwidth starvation for the application traffic on that link. This is also not financially prudent as the enterprise ends up incurring costs on bandwidth that may be sparingly used.
Any kind of intelligent traffic steering decisions that need to be taken for catering to higher bandwidth needs for applications, or use of alternate links because of better quality metrics, is always accompanied by complex manual configurations and interventions at remote locations, where in network IT is almost absent. Cost and manual network management have always been a challenge to enterprises and their network IT teams. IT teams have tried automation as well to better manage these remote locations and links. However, these methods have also fallen short given the overall lack of support from the classical networking devices at the edge.
Dynamic Path Selection with SD-WAN!
Dynamic Path Selection is an algorithmic approach to steer traffic on the available WAN links based on the quality, congestion and utilization metrics. It also provides the Enterprises an option to use multiple underlays available at a branch and combine all the links to provide the best quality of network experience to the user. One can combine MPLS, Internet Leased Line, Broadband, or 4G and use complete Bandwidth in an efficient manner without any manual intervention or worrying about link quality. Path selection algorithms are independent of physical transport or link type (MPLS, ILL, Broadband, 4G) to provide the flexibility of selecting the best available link type available at remote locations.
Enter ScaleAOn Dynamic Path Selection. ScaleAOn Dynamic Path Selection is an innovative technique developed to mitigate the problems like availability and performance in the last mile link connectivity. ScaleAOn Dynamic Path Selection increases network performance and resiliency by harnessing all available WAN transport at a branch, be it MPLS circuits from any service provider or Internet Leased Line, Wired Broadband, Wireless Broadband [3G or 4G/LTE] into a single logical pipe. The system actively monitors the quality in terms of latency, packet loss and jitter, and the capacity in terms of throughput, of the WAN links and steers traffic to the most appropriate link based on the business intent defined as a policy. It combines the power of metric based WAN link selection to steer the application traffic on the WAN link with metrics with an ability to utilize all available WAN links to load balance the application traffic if the metrics of all WAN links are within permissible thresholds.
Metric based WAN Link Selection for Application traffic
The system automatically anchors the application flows with the highest priority on the WAN link with the best quality score. The application flows get progressively anchored across all available WAN links using the quality score as a metric. ScaleAOn Quality Score is a normalized metric that captures end-to-end path roundtrip delay, jitter and packet loss on a WAN link. The Quality score is calculated for each path from one node to another node. The range of this metric is from 0 to 100, where higher value denotes lesser delay, jitter and packet loss. Hence, better the ScaleAOn Quality Score for a WAN link, the better is its quality. The ScaleAOn Quality Score range is further subdivided into Quality Score Bands. The application traffic is steered on the WAN link in the best ScaleAOn Quality Score Band.
Packet by Packet Application traffic Load Balancing
The WAN link selection algorithms often encounter scenarios wherein the link metrics in the form of Link Quality (Link Latency, Jitter and Packet Loss Ratio), Congestion and Utilization for all WAN links at a location are within permissible thresholds. This leads to a scenario wherein all WAN Links connected at the location are suitable candidates to anchor the application traffic. Under such a scenario, the ScaleAOn Dynamic Path Selection Intelligent Engine anchors the traffic across all WAN links by doing a per packet by packet load balancing of application traffic on all the WAN links connected at the location. If there are multiple WAN links whose metrics fall within the best Quality Score Band, then the application traffic is load balanced across all these WAN on a per packet by packet basis. The system does an automatic Packet Order Correction and Random Early Detection (RED) based automatic queue management (AQM) for the interface queue scheduler. This ensures the best quality of experience for the application traffic user. The algorithm automatically falls back to flow based Load balancing after detecting the performance degradation because of huge delays in packet reordering due to large delta of WAN characteristics (Latency, PLR, and Jitter) of multiple WAN links.
Symmetric and Asymmetric Traffic Steering
The system supports asymmetric traffic steering on the forward and reverse path for the Enterprise traffic between spoke and the hub. However, for certain types of applications where the traffic is steered on the WAN path using policy-based WAN path selection, the system intelligently picks a symmetric path for the traffic between the hub and spoke sites.
The combination of metric based WAN link selection and Packet by Packet load balancing allows for a superior network for the application traffic leading to better Quality of Experience for application users. The application traffic gets the best network even under the stressed WAN conditions at the branch, with rapid response to changing WAN conditions like WAN Quality changes and Link Flaps.
The application of the ScaleAOn Dynamic Path Selection is agnostic to the WAN link type at the remote location and supports all available WAN transports in the market today. Lavelle Networks’ ScaleAOn Dynamic Path Selection is policy driven and intelligent, requiring no manual intervention at the edge device. The policy definition is centralized in the CloudStation controller while the policy enforcement happens at every device centrally through the controller.
The ScaleAOn Dynamic Path Selection ensures that application traffic always gets the best WAN path(s), with no impact to the Quality of Experience because of one bad WAN link at the remote location. The ScaleAon Dynamic Path Selection is based on the fact that there are a multitude of service providers, some local and some global, who can serve any given enterprise location, anywhere in the world. It also utilizes the fact that Internet bandwidth is inexpensive and provides the ability to build a scalable and always on Enterprise WAN connecting physical and virtual locations of an Enterprise anywhere in the world.