Cloud services hinge on application performance on WANs
Cloud computing is becoming pervasive and cloud-delivered hosted applications are presenting new challenges ...
Cloud service providers and their enterprise customers mutually can benefit from rock-solid wide area network performance.
By Rob Marson, JDSU
Cloud computing is becoming pervasive and cloud-delivered hosted applications are presenting new challenges and business opportunities. Today, network operators are under constant pressure to reduce costs, maximize profits, guarantee quality of service, increase customer satisfaction and support new and real-time applications—all while balancing ever-increasing bandwidth demands. Network operators, however, play a critical role by connecting users to applications and data. Strategically situated between downstream customers (typically enterprise or business clients) and upstream customers (typically application service providers delivering hosted solutions from a data center), network operators deliver a variety of connectivity solutions using metro Ethernet and Internet Protocol (IP) virtual private networks (VPNs), as well as a variety of private line services, to provide the onramps to the cloud.
Even so, today’s application delivery chain is somewhat decoupled. Network operators typically provide only a basic level of service level agreements (SLAs) to upstream and downstream customers, based on data delivery, not application performance. The wide area delivery network has little to no visibility to the transactions traversing it, yet its performance is critical to the end-user experience. Slight changes in network performance may have significant consequences to application performance.
|Applications using TCP/IP can be significantly affected by packet loss and application latency. Identifying the location and sources of unexpected latency is paramount for a cloud-delivered service provider to offer meaningful SLAs.|
The adoption of applications and technologies such as video, Voice over IP and virtual desktop infrastructure (VDI), as well as different flavors of unified communications (UC) technologies, are increasing the complexity of wide area network (WAN) traffic and posing new challenges for managing WAN performance. Multiple WAN technologies may be involved; customers may reside in large enterprise locations or small/remote offices, while others rely on mobile devices to access their applications and data. This non-heterogeneous nature of application delivery further complicates the ability to monitor application performance or to offer any application for SLAs.
WAN’s critical role
The impact of poor network performance on overall application performance can be significant. Packet loss can occur for many reasons including network congestion, protection events such as route reconvergence, or network misconfigurations. Furthermore, most WANs are oversubscribing, leveraging statistical gain techniques to maximize the use of available capacity. As data traffic is bursty and unpredictable in nature, the wide area switches will begin to selectively discard traffic when network congestion occurs. The percentage of packets discarded may be low—perhaps in the region of 1 to 2 percent—but the impact on application performance can be dramatic.
While this is acceptable to some applications that do not require real-time interactive performance, many critical applications such as those necessary for data center virtualization and hosting are intolerant of packet loss. This often results in drastic application decreases in throughput as network packet loss increases, and is further compounded by a significant increase in latency. Within the network, identifying the location and sources of unexpected latency, as well as the associated implications to specific applications, is paramount to be able to offer meaningful SLAs to cloud-delivered services. The effect of packet loss and application latency for applications using TCP/IP is shown in the chart within this article.
The situation today
When it comes to having full visibility into network traffic, it is no longer sufficient just to monitor bandwidth consumption per protocol (i.e. HTTP, SSL, etc.) to understand how much bandwidth is being consumed by each application. For example, not all HTTP traffic has the same level of business priority or bandwidth requirements and should not be treated as a homogeneous category. Without having a sufficient level of application-level visibility, it is difficult for organizations to ensure that their available network resources are being used in a way that provides the most value to their businesses.
Today’s network equipment, the switches and routers, which work together to form the WAN, are unaware of the applications and transactions that traverse them. Today network operators have no simple way to measure user application response times in order to understand and manage user experiences. Similarly, there is no correlation available to understand how WAN infrastructure changes may affect network and application performance. Operators are unable to isolate performance problems to the application, server or network.
Today’s SLAs focus on the physical layer to the networking layer (Layers 1 to 3). Basic performance monitoring reported includes latency, packet delivery ratios, and errored seconds. Distributed applications require distributed data-capture strategies in order to be able to isolate issues. Current solutions usually require appliances to capture and monitor network traffic, including specific applications. Extending this visibility to all different locations, however, is complex and costly. While in a data center a typical large appliance—like a network recorder capable of capturing data at 10 Gbits/sec by monitoring critical links—may be a viable strategy, the approach does not scale as the WAN extends to enterprise campus locations and small offices, or branch offices where the cost to deploy and sustain a hardware appliance becomes prohibitive. Obtaining a true measure of user experience requires in-line monitoring at client and server end points, as well as throughout the dynamic WAN.
|Shown on top is a PacketPortal SFP probe from JDSU, which is a key element in the company’s PacketPortal system. The probes can be installed anywhere in a network; PacketPortal was designed to provide the data- and intelligence-capture capabilities described in this article.|
An effective strategy requires the ability to look beyond packet headers and deep into the content within the applications. For instance, the growing mobile user presence and the increased complexity of network environments means that users typically pick up IP addresses dynamically. For this reason, a user could have several IP addresses during a single work period. This evolution of the network makes it nearly impossible to monitor, secure and manage solely by an IP address. Unless a simple, pervasive and cost-effective deep packet inspection capability is available, it will be difficult to monitor exactly what traffic may be traversing networks.
Business opportunities for service providers
Providing enhanced WAN traffic visibility creates opportunities for service providers to offer advanced SLA-based solutions that process data collected and make it more actionable for their own use as well as their customers. Many enterprises lack the capital budget to deploy a robust WAN monitoring product while others lack the in-house expertise to get the most out of the technology. In these situations, an enterprise can improve network visibility and its response to network issues by outsourcing WAN monitoring and management.
By outsourcing WAN monitoring and management to a trusted third party, an enterprise can eliminate the hardware maintenance and support staff costs associated with having on-premises technology. The managed service provider can also take advantage of economies of scale to use the most sophisticated WAN monitoring and management products on the market and employ people with the expertise to assess more effectively the data collected. This skill set qualifies a carrier to resolve any trouble spots, accurately forecast future capacity requirements and identify possible denial of service attacks before they wreak havoc. These services can also be bundled in with transport and other managed services. Small- and medium-sized businesses with fewer IT personnel are ideal candidates for outsourced WAN performance management services.
In order to deliver these types of services, network operators need to have visibility into how their network capacity is being used, how applications on the network are performing, where the bottlenecks for optimal performance are, and be able to make educated decisions about how to improve network and application performance.
Value on all sides
Examples provided by Google show that an extra 0.5-second delay in generating search results would worsen the user experience and, in effect, reduce traffic to its website by 20 percent. Clearly there is untapped potential for operators to improve the quality of the end-user experience on behalf of content providers. Proactive and real-time monitoring, and quickly resolving any issues that lead directly to consumers not completing online purchases, abandoning video-on-demand services, or otherwise switching to other content services achieve this.
From a cost-containment perspective, operators may use increased intelligence about utilization on their network to better plan capital investments as well as develop new business models with over-the-top (OTT) providers. For example, in a 2011 company filing, Dutch network operator KPN has cited OTT applications as a reason for an 8.1-percent drop in domestic service revenue. Similarly, for a mobile provider, VoIP influences revenues from cellular roaming. With VoIP, a user can talk for more than 22,000 minutes before exceeding a 5-Gbit data cap. Knowing exactly how the network is being used could allow the operator to create advanced OTT SLAs, improving the content delivery for upstream OTT partners, enriching the user experience, all while maximizing revenues from SLA-backed services. Furthermore, network operators can use this insight to better and more strategically plan future capital investments.
Requirements for new solutions
By leveraging a network monitoring solution providing access to key application-based performance information, network operators can provide advanced SLAs to their upstream (e.g. application service provider) and downstream (e.g. business user) customers. These may include the following.
- Measurement and analysis application performance for all transactions
- Comparisons of response times against intelligent baselines and thresholds
- Identification of abnormal latencies in the network
- Isolation of the problem to a specific link or application server, or the application itself
- Delivery of alerts on any performance deterioration
Smart network application platforms provide a cloud-based approach to embed data-capture technology throughout the network, delivering inline intelligence to any monitoring, management or business application. This provides unprecedented access to intelligence across the network and especially to the network edge, where 80 percent of customer issues occur—making networks more efficient and cost effective. Most importantly, cloud-based networks preserve the value of existing infrastructure and personnel while enabling these resources to perform better and more efficiently.
Rob Marson is a strategic marketing manager with JDSU’s CommTest division (www.jdsu.com), serving market segments that include carrier Ethernet, optical and cloud networking.