An Interview with Karim Fodil-Lemelin, Xiplink Sr. V.P. of Engineering...

With Billions at Stake, Can LEOs Meet the FCCs Latency Hurdle?

In the race to capture a share of the FCC's $16 billion Rural Digital Opportunity Fund, latency is key. While the FCC agreed to allow low-Earth orbit satellite companies to apply for funds as low-latency broadband providers, it expressed doubt that any of the LEO providers could meet the sub-100 milliseconds latency requirement.

For SpaceX's Starlink constellation, crossing the 100 ms threshold is critical. A share of the billions of dollars in federal money would go a long way toward funding the multi-billion dollar cost of its LEO Constellation.

Unsurprisingly, SpaceX has claimed that Starlink can easily meet the 100 ms requirement including, "processing time." Even further, Elon Musk stated that latency could easily be as low as 10 to 20 ms - claims doubted by the FCC.

While everyone in the satellite industry understands that uplink and downlink times to a LEO are a fraction of those to a GEO, few are aware of the potentially excessive delays encountered in terrestrial network transmission. To understand the challenges Musk and other LEO operators face in meeting the 100 ms latency requirement, we met with Xiplink Sr. V.P. of Technology, Karim Fodil-Lemelin. 

For those unfamiliar with XipLink, the company is the premier authority on satellite network optimization. It's products are widely deployed in the satellite industry by such well- known companies such as Carnival Cruise Lines, ST Engineering, Intelsat, SES and others.

SMW: Karim, As you know, LEOs have significantly lower latency due to their greater proximity to the earth. However, latency over a satellite connection is always significantly more than the simple round-trip time between the earth station and satellite. Can you explain why?

Karim Fodil-Lemelin (KF): While latency in a LEO constellation is less in a GEO satellite link, physically reducing the communication distance is only one factor in the overall latency. There are many other factors inherent in the totality of any satellite or terrestrial network that can result in significant delays.

Whether a LEO or GEO constellation, every step along the way, the IP stream encounters various routers, firewall, and intrusion detection systems, optimizers, modems, and devices that may act as bottlenecks, sending and receiving data slower than the actual satellite link. 

In each of these devices, the data stream may be queued to manage the burstiness inherent in IP transmission. In a typical network, the IP stream may have to go through as many as 20 to 30 of these devices, each potentially queuing data. 

As congestion across the network increases, so does the queuing and so the latency. In the worst case, network congestion can overload the queues in the devices, resulting in packet loss and further delay in the overall transfers. All these factors potentially increase the actual latency far beyond the latency inherent in the satellite link itself. re-transmission results in additional delay for the end-user. The key to minimizing re-transmissions is to avoid network congestion. 

"As congestion across the network increases, so does the queuing and so the latency. In the worst case, network congestion can overload the queues in the devices, resulting in packet loss and further delay in the overall transfers. All these factors potentially increase the actual latency far beyond the latency inherent in the satellite link itself."

SMW: What are the consequences of Packet Loss?

KF: If packets are lost, at the other end of the communication, the receiver will see that it did not get the packet and ask for it to be resent. The command to resend the packet as well as its re-transmission results in additional delay for the end-user. The key to minimizing re-transmissions is to avoid network congestion.

SMW: What is the major cause of network congestion?

"Typically, it's a bandwidth issue. At some point in the network, a bottleneck can form, resulting from more traffic being pushed down the ?pipe? than the "pipe' allows. So, insufficient bandwidth leads to over queuing and ultimately packet drops. Because there is no such thing as an infinite bandwidth network, network congestion needs to resolved. 

SMW: How do you detect congestion in the network, and what can be done to alleviate it? 

KF: The first step in alleviating congestion is to measure network latency. As it increases, it?s an indication that network traffic is backing up in the queues. To detect it, we insert a Performance Enhancement Proxy, our Wireless Link Optimizer (WLO), into the network, ideally near the bottleneck. Using Delay Based Rate

In each of these devices, the data stream may be queued to manage the burstiness inherent in IP transmission. In a typical network, the IP stream may have to go through as many as 20 to 30 of these devices, each potentially queuing data.

As congestion across the network increases, so does the queuing and so the latency. In the worst case, network congestion can overload the queues in the devices, resulting in packet loss and further delay in the overall transfers. All these factors potentially increase the actual latency far beyond the latency inherent in the satellite link itself.

SMW: What are the consequences of Packet Loss?

KF: If packets are lost, at the other end of the communication, the receiver will see that it did not get the packet and ask for it to be resent. The command to resend the packet as well as its re-transmission results in additional delay for the end-user. The key to minimizing re-transmissions is to avoid network congestion.

SMW: What is the major cause of network congestion?

Typically, it?s a bandwidth issue. At some point in the network, a bottleneck can form, resulting from more traffic being pushed down the ?pipe? than the ?pipe? allows. So, insufficient bandwidth leads to over queuing and ultimately packet drops. Because there is no such thing as an infinite bandwidth network, network congestion needs to resolved.

SMW: How do you detect congestion in the network, and what can be done to alleviate it?

KF: The first step in alleviating congestion is to measure network latency. As it increases, it?s an indication that network traffic is backing up in the queues. To detect it, we insert a Performance Enhancement Proxy, our Wireless Link Optimizer (WLO), into the network, ideally near the bottleneck. Using Delay Based Rate Control, the WLO will manage upstream congestion, and, if necessary, initiate a request to slow down the sending rate. In addition to latency detection and active queue management, the WLO has several other capabilities that speed up the network and reduce congestion.

When possible, it compresses and byte caches data, virtually increasing bandwidth and reducing the need to re-transmit data that has already been transmitted.

Because the WLO can often detect packet loss before it happens, the data transmission can reach its destination point without interruption. If the loss is unavoidable, the WLO can masquerade as the end point and order re-transmission of the damaged packets, or in some instances, it can re-transmit the damaged packets on its own. That saves bandwidth in parts of the network and eliminates the round trip time to and from the data endpoint.

Lastly, if congestion becomes unmanageable, as the last resort, the WLO can selectively drop packets, thereby easing network congestion.

XipLink is focused on ?goodput?, meaning useful customer data (not re-transmits), resulting in link utilization rate increases from less than 50% in extremely chatty situations like secure links to 98% in a typical deployment. Incrementally, more bandwidth is also provided above 100% utilization when compression, caching and other WOL techniques are implemented.

As a result, users typically benefit from 30% or better reduction in response time, resulting in a significantly better user experience. 

The XipLink device itself can be incorporated into the network as a hardware appliance, or virtualized. It can be placed at the data center and optimize the entire network, or it can isolate the satellite link through the placement of WLOs at the satellite uplink and downlink.

The combination of features offered by the XipLink Wireless Network Optimizer is invaluable in reducing latency, maintaining user sessions, and avoiding packet loss over satellite and other networks.

If LEO operators want to meet the FCC's latency requirements and secure a portion of the funds allocated for rural broadband, the Xiplink Wireless Network Optimizer could help them clear the 100 ms barrier.

About Karim..

Mr. Karim Fodil-Lemelin is responsible for the architectural conception and implementation of the XipOS operating system, which is at the heart of XipLink's products.

In telecommunications since his graduation from Physics Engineering at Polytechnic school at University of Montreal, he led several research projects for the Canadian government in the field of IP communications over satellite links.