Infineta Systems Shakes Up the WAN Optimization Landscape

The case for WAN optimization to accelerate data movement across the WAN has been clear for some time now. Over the last 7-8 years, the market has grown large enough to support a number of players. WAN optimization vendors have been progressively evolving their product offerings by delivering incremental performance improvements in their latest and greatest devices. These vendors have tended to focus more on the branch to data center application performance problem and less so on the inter-data center WAN.

Infineta Systems, a San Jose-based maker and seller of what it terms ‘Big Traffic WAN Optimization’ solutions, seems to have leap-frogged the traditional WAN optimization players in the inter-data center space. Infineta announced the general availability of its Data Mobility Switch (DMS) in June 2011. The DMS seems to be the only WAN optimization solution in the market that scales to 10Gbps speeds, compared to 2.5Gbps available from the nearest competitive product.

Branch WANs and inter-data center WANs are quite dissimilar, and the difference isn’t just the typical bandwidth rates on each. Branch office WANs tend to be measured in the 10s of Mbps and support user-to-machine type of applications such as file access, email and web-based applications, driving a number of low-throughput flows. Inter-data center WANs tend to be measured in Gbps and support machine-to-machine type applications such as replication, backup and large file transfers, driving fewer, high-throughput flows (100Mbps to 1Gbps) that may potentially be established for very long periods of time.

Given the scale of traditional WAN optimization solutions, many enterprises with multi-Gigabit links between their WAN links tend to walk away from the incumbent solutions because no solution out there scales to the levels at which enterprise data centers WANs are operating today. Instead, enterprises have tended to upgrade their bandwidth or relax their data protection goals, both sub-optimal choices. As anyone who has deployed replication traffic over the WAN knows, adding bandwidth does nothing if the WAN latency is high enough. Relaxing data protection goals (e.g. allowing RPOs to grow or backup windows to get longer) is career suicide. But the rate of traffic traversing inter-data center WAN links isn’t letting up. Particularly with enterprises exploring data-intensive IT strategies such as virtualization and Big Data, inter-data center WANs will continue to be the weakest link.

So why does Infineta think it can solve the inter-data center WAN problem where others have failed? Well, they say it’s their choice of merchant silicon, a networking performance based approach, and their patented deduplication algorithm that makes the difference. As the Infineta team explained to me, traditional WAN optimization solutions cannot scale to the multi-Gigabit scales needs for today’s inter-data center WANs because they are essentially software implementations running on a standard x86-based server. Operations such as network deduplication are extremely CPU-intensive and are one of the larger bottlenecks that the incumbents haven’t been able to address to date. As validated by ESG in their recently published test report, Infineta’s DMS not only scale to 10Gbps speeds, but delivers extremely high reduction ratios at such high speeds, and can maintain high-throughput flows at round-trip times of well over 100ms.

The DMS architecture is quite similar to the approach taken by many of the high-performance networking solutions out there. The DMS is designed around a non-blocking switch plane, uses a number of network processors for line-rate packet processing, and utilizes programmable logic for high-efficacy network deduplication.

The DMS data reduction pipeline is made up of three core stages:

• Repeating Byte Suppression – A key stage in Infineta’s reduction pipeline, Repeating Byte Suppression (RBS) identifies and removes repeating byte sequences from incoming data at up to 10Gbps rates.
• Velocity Dedupe Engineâ„¢ – Implemented in programmable logic, the Velocity Dedupe Enginetakes a distinctly novel and groundbreaking approach to identify and remove redundancies in network traffic. Some key highlights are:
  • Carrying out true byte-level pattern identification at line rate (up to 10Gbps) without slowing down flows. Even the smallest of changes in inter-data center traffic are identified in order to deliver the highest levels of WAN traffic reduction.
  • Performing in-place updates of repeating byte patterns in the deduplication dictionary, resulting in an optimally utilized system.
  • Executing all functions in a massively parallel fashion, the core reason for the high data rates that Infineta’s solutions can achieve.
• Standards-based Compression – The final stage of the reduction pipeline, compression is implemented in hardware and helps to further reduce traffic already reduced by RBS and the Velocity Dedupe Engine.

The DMS TCP optimization engine (known as the Velocity Transport Engineâ„¢) provides the following functions:

• Dynamic WAN resource allocation – Highly responsive processing engines perform prioritized reallocation of WAN resources across all active TCP connections without impacting critical traffic. Relative fairness is achieved quickly and effectively.
• Rapid start and recovery support – By keeping a dynamically updated, end-to-end view of the WAN link and competing flows, the system can enable premium (aggressive) flows to instantly establish and take up large amounts of bandwidth as needed. Connections for workflows such as replication and migration are assured 1Gbps or higher throughput if needed, even in the face of high latency and packet loss. Non-premium connections are also protected and are assured lower but sustained rates as per configuration.

So can the traditional WAN optimization players catch up to Infineta’s merchant silicon-based bullet train by hanging on to Intel’s coattails along the Moore’s law curve? According to Infineta, when the competitors get to 5Gbps, it will be at 20Gbps. That’s quite a claim, making this a fascinating race to watch!