28 Jackson Blvd Ste 1020 Chicago IL 60604-2340
+1 800 559 6580

Packet Loss Recovery

Modern TCP stacks have become more efficient at dealing with latency by adding TCP Window Scaling and Selective Acknowledgements. However, TCP is still vulnerable to packet loss and this has a drastic impact on network performance. Some degree of packet loss is almost always present in every wide area network. At times it is due to link quality but at other times packet loss is intended. This is when core routers avoid congestion and buffer bloat by randomly dropping some packets when the queues grow too large. Where WAN or VPN performance is limited by the negative effects of packet loss and retransmissions, Packet Loss Recovery can be configured.

Why is Packet Loss important?

  • A loss has a severe impact on TCP performance: For example, TCP throughput over a 100 Mbps link is limited to 1 Mbps at 80 ms and 2% loss.
  • More businesses are connecting their offices via low-cost VPN links. These typically have 1% loss rate, which is 10 times higher than MPLS links.
  • SLA’s provide a false sense of security. A 3+% loss for an hour a day means providers can still meet a very low monthly average loss rate SLA.
  • Packet Loss is higher in emerging and developing countries. On average 2% or more.

How Packet Loss Recovery works

Enixea provides three layers of loss recovery.

  • Fast Retransmit Replay: Since version 2.6 loss recovery worked by monitoring packet sequences and using a real-time round trip time of the link between two Enixea peers. When a packet is lost, a timer expires and the packet is retransmitted. Packer Order Correction ensures the end hosts are unaware of the losses on the WAN and the TCP layer does not reduce the transfer rate. Below is a chart of v.2.6 PLR performance.
  • Fast Retransmit Request: Since version 4.0 an optimization is added to further increase Packet Loss Recovery acceleration. Fast Request works by monitoring packet sequences and when a lost packet is detected a replay request is sent to the peer. In this case, the lost packet can be detected faster and replayed sooner, increasing the acceleration benefits.
  • Forward Error Correction: Since version 4.1 FEC can be configured. In the case of FEC, a special packet is sent to every group of 5, 10 or 20 packets. If a packet in the group is lost, the lost packet will be reconstructed in real time. FEC adds an additional acceleration benefit in conjunction with other PLR and TCP Acceleration features. FEC is an optional feature that can be enabled when required.

Packet Loss Recovery Results

To counter the negative effects of loss on TCP throughput, Enixea can be configured to provide packet loss recovery to assist in TCP Acceleration. Below are CIFS copy speed stats over various loss and delay profiles. The first control test indicates the standard TCP throughput under these conditions. The second test is with Enixea Packet Loss Recovery enabled and compression and deduplication disabled.

Packet Loss Recovery Performance

Packet Loss Recovery Performance

 

Configuration

From version 4 onward PLR is auto-configured. For configuration of earlier versions, please see the following guide: Configuring Packet Loss Recovery

 

2 Responses

  1. admin
    Our Enixea VMs is running wonderfully. Not a single hiccup on any of the four that we are running. Just chugging along without incident for months without a reboot. Very impressive.
    • admin
      With no previous experience, we were initially a bit cautious with the Enixea Wan Optimization solution proposed by Progility. After a short trial for the cost, we were amazed how well the solution works and how seamlessly the solution incorporates into our existing infrastructure. We can’t believe it didn’t appear on our radar years ago

Leave a comment