WAN Acceleration: A Virtual “Must Have”

Optimizing Application Performance and Disaster Recovery in a Virtual Server Environment

By Jeff Aaron, Director, Product Marketing, Silver Peak Systems

In a virtualized environment, IT managers need to pay careful attention to the impact that Wide Area Network (WAN) performance has on application performance. When virtual servers are placed in centralized locations, limited bandwidth, high latency, and packet loss on the WAN can impact application performance for end users. At the same time, the WAN can present a major obstacle for data protection and disaster recovery in these environments when large virtual images must be replicated between geographically disperse locations. For all of these reasons, WAN optimization has become a key enabler for strategic virtualization projects.

This paper will discuss in more detail the challenges of doing virtualization across the WAN, and identify common WAN optimization techniques that improve virtual application performance while improving the backup and recovery of virtual machines.

Server Consolidation

There are many reasons why enterprises turn to virtualization as a way of consolidating application servers and databases. While hardware and management costs are typically the most recognized, performance, scalability, and security benefits can also be primary drivers.

Virtual machines suffer all the same performance challenges as physical servers when accessed across a WAN. More specifically, the following WAN characteristics can all adversely impact the performance of centrally hosted virtual applications:

  • Limited bandwidth. Depending on the volume of data being accessed and transferred across the WAN, bandwidth can be a huge concern in a virtualized environment. WAN speeds typically function at a fraction of LAN speeds, which creates a natural bottleneck that can adversely impact the performance of many virtual applications.
  • High Latency. You cannot break the laws of physics. It takes time to physically communicate from one location to one another, which can be further exacerbated by “chatty” communication protocols, like the Transport Control Protocol (TCP). The impact that latency will have on performance will depend on the type of application being hosted centrally.
  • Packet loss. As enterprises move increasingly to MultiProtocol Label Switching (MPLS) and Internet Protocol Virtual Private Neworks (IP VPNs), packet loss is becoming a bigger and bigger problem. These networks are oversubscribed by the carriers, which can result in packets being dropped or delivered out of order during times of heavy congestion. Packet delivery issues such as these are especially problematic when high data volumes must be sustained across the WAN.
Disaster Recovery
Virtualization technology can make it easier and more cost effective to implement disaster recovery. Instead of requiring a 1:1 mapping between physical hosts and targets, which effectively doubles infrastructure costs, virtualization allows a single physical server to act as a recovery point for many virtual machines. This limits the amount of hardware required for data protection and recovery. In addition, this simplifies the disaster recovery process by eliminating the need to manage disparate servers with disparate operating systems.

VMware and other virtual solutions have snapshot capabilities to regularly replicate changes to target virtual machines. In the event of an outage, the replicated virtual machine can be started as a backup device with the most recent data. In addition, numerous 3rd party solutions exist that provide real-time replication of virtual machines to target devices for maximum data protection.

Both the replication and snapshot processes can generate an enormous amount of traffic, which can create a challenge when the process is taking place across the WAN. In addition, both leverage TCP for transport, which can create latency and cause backup tasks from being completed in allocated windows. This results in database synchronization issues and missed Recovery Point Objectives (RPO). Lastly, many replication processes require high sustained data throughput, which cannot be disrupted until the entire replication process is complete. If the flow of data is disturbed, as can occur if packets are dropped or delivered out of order across the WAN, effective throughput across the WAN will never exceed 1 or 2 Mbps regardless of how much bandwidth is actually available. This will bring a replication process to its knees.

WAN Acceleration: A Virtual “Must Have”

WAN acceleration addresses the common bandwidth, latency and loss issues that can hamper server centralization and data protection plans, making it an essential component for enterprise virtualization. More specifically, WAN acceleration provides the following benefits in virtual environments:
  • Improve data transfer times: WAN deduplication is a new technology that has moved to the forefront of the WAN acceleration space. It works by delivering duplicate data from local data stores instead of resending it across the WAN. WAN deduplication can have an enormous impact on data transfer times, which means better perceived performance for virtual applications. In addition, faster data transfers improve the performance and reliability of replication/recovery processes.

Figure 1: WAN deduplication improves data transfer times and maximizes WAN efficiency by eliminating the transfer of duplicate data across the WAN

  • Maximize WAN efficiency: WAN deduplication can reduce as much as 99% of WAN traffic by eliminating the transfer of duplicate information. With byte level granularity, repetitive patterns can be detected within a single transfer, across separate transfers, and across different virtual applications. In this respect, WAN deduplication complements deduplication that might already be taking place in the host or replication software.
  • In addition, advanced header and payload compression techniques can be used to reduce the amount of WAN bandwidth consumed by virtual applications when accessed across the WAN- even when the information is not repetitive.
  • Reduce packet loss and errors. WAN acceleration can be used to reduce the impact of both packet loss and jitter that occurs when router links are oversubscribed and drop or re-order packets (as is common with shared IP networks, such as MPLS and IP VPNs). Adaptive Forward Error Correction (FEC) for example, rebuilds lost packets on the far end of a wan link in real-time, while Packet Order Correction (POC) reorders packet in real-time. Both techniques eliminate the need for re-transmission, which can lead to poor application performance and failed replication processes.
  • Increase geographic distances. By reducing the impact of latency, enterprises can extend the distances between users and data, enabling virtual servers to be located anywhere in the world (and backed up to disaster recovery locations located anywhere in the world.
  • Protect virtual traffic. Many WAN acceleration devices use encryption to protect network traffic sent across the WAN. This adds an added element of security to those enterprises concerned about unauthorized access to data stored on virtual machines.

Just as bandwidth, latency, and loss can hamper the performance of applications running on physical servers, the same is true of virtual servers. In addition, just as these WAN challenges can hamper backup/replication processes between physical hosts, they can also impact the performance and reliability of data protection in a virtual environment. As a result, WAN acceleration is strategic to many server virtualization initiatives.

Silver Peak Systems Ltd is exhibiting at Storage Expo 2008 the UK’s definitive event for data storage, information and content management. Now in its 8th year, the show features a comprehensive FREE education programme and over 100 exhibitors at the National Hall, Olympia, London from 15 - 16 October 2008 www.storage-expo.com

Source: Eskenzi PR Ltd