The core VXOA technology uses a variety of techniques to reduce latency and improve throughput over WANs, including network memory, latency mitigation and forward error correction, says Damon Ennis, Silver Peak VP of product development.
VXOA uses network memory to reduce the amount of traffic sent between two computers on the network, which can reduce the quantity of data sent between applications by five to 20 times, says Ennis. In some cases, this may be the result of padding incorporated into certain file types. In most cases, this is the result of inefficiencies between the application and the TCP/IP protocol stack.
VXOA monitors traffic at the IP layer to identify repeating data patterns at the packet layer in a process called fingerprinting. When a VXOA sender identifies a reused pattern, it creates a reference to the data, which can be used by the receiving system to re-create the data.
According to Ennis, the approach is distinct from others used in WAN optimization in that it operates lower on the protocol stack on IP packets and is less brittle than other tools that deduplicate data at the application layer. Consequently, the VXOA infrastructure can support new applications without the need for special plug-ins.
On the latency mitigation front, the VXOA platform uses an intelligent feedback mechanism to improve TCP protocol window sizes. Normally, TCP increases the number of packets sent in each cycle, until the system detects that some packets are being lost or delayed. Ennis says this approach doesn't work as efficiently on high-throughput networks because it takes more time to ramp up the transmission speed to the full capacity of the network and slows transmission down below the maximum available capacity. VXOA is able to track network congestion and throughput using a variety of probes, and can dynamically adjust TCP window sizes to take advantage of maximum network capacity.
A third challenge is that WAN networks tend to lose packets more frequently than LAN networks because of network contention caused by a larger number of shared users. Even a small amount of packet loss can have significant performance issues for high-performance networks because the receiver has to wait for all required packets to reassemble the application data. When a VXOA system detects packet loss, it automatically begins adding redundant data in a process known as forward error corrections. This allows the receiver to recreate lost packets, without waiting for a packet retransmission to occur.
VXOA 5.2 can be deployed as a dedicated hardware appliance (Silver Peak NX-10K) or a virtual machine (Silver Peak VRX) running on a VM server. The VM version could previously be installed on VMware ESX and Microsoft Hyper-V. The new release also adds support for Xen and KVM hypervisors, which should make the technology more accessible for organizations that have standardized on these lower-cost VM platforms.
VXOA 5.2 adds support for a larger number of TCP flows, as well, which is particularly important for cloud-based applications with a need for lots of simultaneous connections, says Ennis. The hardware version now supports 512,000 flows, while the VM version supports 256,000 flows.
One of the challenges with moving virtual machines between physical servers is the need for Layer 2 Ethernet connectivity. Router vendors have developed protocols for creating LAN tunnels for these connections over a wide area such as Cisco's Overlay Transport Virtualization and UDP-based Data Transfer Protocol. The third major improvement in VXOA 5.2 is support for automatically optimizing these tunnel protocols.