Packet Reordering

Modern network stacks#

Most modern network stacks are designed for application access from a wide variety of locations and devices, often involving global or regional reach.

As a result, a small amount of out of order packets is inherently coped with by the TCP stack implemenation on the servers hosting the applications, or in the applications themselves, to cope with changing last mile network conditions.

Linux has the besta and most standards driven TCP stack and so rarely experiences any issues.

The newest Windows Server implementations have been upgraded improving the way they handle out of order packets.

Aggregation and reordering#

Due to the nature of packet distribution across differing latency links, there can be some out of order packets. As indicated above, modern TCP stacks have no issue with this and do not see any degredation in application performance.

Where legacy TCP stacks and applications are in use, dedicated packet reordering may be required.

Microsoft TCP stack#

Microsoft have been slowly rolling out improvements to their TCP stack which eliminate the need for reordering. The newest application infrastructure and versions of Windows Server have the latest TCP stack, but older applications and hosting services can experience slow downs in bandwidth usage.

For these applications we would recommend reordering that specific traffic.

1 Gbit/s data streams#

Our packet reodering engine can reorder up to 1Gbit/s per data stream. In practice this is in line with most user and client requirements.

Load balancing multiple 1 Gbit/s links can sometimes be preferable to reordering individual streams.

Note this is NOT Link Aggregation (LAG) and would would still benefit from all the other network features including IP affinity, QoS and complex network presentaiton designs, as the load balanced traffic is still encapsulated within the Tunnel Aggregation Group.