What does ISP ping really mean?

Question

I saw an article detailing Starlink's upcoming preorders and they mentioned that Starlink will have "latency from 20-40ms". Knowing even the basic definition of "ping" made me immediately question what that figure was supposed to represent.

Ping, otherwise known as latency, measures how long it takes to communicate with another server and receive a response back. This depends on a number of factors:

1. Quality of underlaying infrastructure (wireless, optical, copper, etc.)
2. Network routing efficiency
3. Physical distance

In my experience with terrestrial ISPs "physical distance" is by far the biggest contributor to this number. Online gaming, for example, gives a similar ping for myself and my co-located friends because we're playing from the same general area (a city) and communicating in the same general area (where the game host servers are).

However, when I perform internet speed tests or see descriptions of services like Starlink - I see a "ping" metric posted as well. From what I know about ping, this is borderline meaningless. The "ping" figure is meaningless without context. Speed tests in particular intentionally target nearby datacenters, so your ping will frequently be extremely low. Starlink is more of an outlier, because nothing is "close" by terrestrial standards (being 300-400 miles up).

To return to the Starlink example, one simple scenario immediately highlights why claiming any kind of latency number is suspect: communicating with other countries. As anyone who's used VoIP or online gaming with others from across the globe knows, 200ms ping times are not uncommon simply because of the distance involved. Starlink, or any ISP, can't claim a latency without establishing some kind of context. It's like selling a car and claiming "this car gets 500 miles to a gallon!" without disclosing that the test course was entirely downhill.

I know it's idealistic, but a far more useful number would be something like "% latency added", where it measured latency differences between theoretical maximum speed (unimpeded light travel) and observed latency. This would inform the consumer how well/poorly their specific network/ISP is performing in a general sense - not just how quickly it can talk to a datacenter 30 miles away. When evaluating specific ISPs, the network latency is just as important as bandwidth (arguably moreso), yet it is misleadingly or inaccurately reported.

Is the "ping" number reported by speedtests or ISPs themselves useful or standardized in any way, or is it purely a marketing gimmick?

Answer 1

You're right that ping times, being round-trip travel times for packets, are meaningless without the context of where the packets are traveling from and to.

For Starlink, it's safe to assume the "from" is "from the user's Starlink terminal at their home".

The "to" is probably somewhere on Starlink's terrestrial network. It could be the near edge, the middle, or the far edge. That is:

• It could be the nearby ground station where the packets enter the terrestrial network from the satellite network. That would help tech analysts compare how Starlink stacks up against competing customer-edge access network technologies like DSL, DOCSIS, and GPON (and of course old school crazy-high-latency geostationary satellite broadband like HughesNet).
• It could be a Starlink data center "in the middle" of Starlink's terrestrial network, where all the cloud providers and CDNs could co-locate edge nodes (including game servers). So in this way, it could in fact be directly representative of the ping time you'd see in-game, at least for game services that choose to colocate edge nodes at Starlink's data centers.
• It could be the far edge where Starlink's terrestrial network connects to Internet backbone providers and peer ISPs. In this case it tells you the minimum ping time you'll ever see when pinging any host on the Internet from your Starlink-connected home.

The way the Question was worded suggests a couple possible misconceptions that deserve to be cleared up:

First, please note that although Starlink has talked about future plans for satellite-to-satellite laser links to create a mesh network of satellites, that part isn't up and running yet, last time I checked. So your VoIP call to the other side of the globe isn't being routed satellite-to-satellite across the mesh of satellites to the other side of the globe. Instead, it's being bounced off a single satellite to a nearby ground station, where it connects to their terrestrial network just like any terrestrial ISP would have. Because of this, Starlink service is only being offered in places surrounding a location where Starlink has been able to build a ground station, because any given packet you transmit must be able to hit a single satellite "in view" in the sky over your house, and that satellite must also be "in view" of a Starlink ground station (a big Starlink-owned dish, not a customer terminal) at the same time. So for now, Starlink is much more like a series of regional broadband access networks, NOT a long-haul Internet backbone that just happens to be orbital.

In my experience with terrestrial ISPs "physical distance" is by far the biggest contributor to [RTT measurements]. [...] As anyone who's used VoIP or online gaming with others from across the globe knows, 200ms ping times are not uncommon simply because of the distance involved.

You're right about "physical distance" and you're right to put it in quotes, because the distance itself isn't the issue, because the signals travel at the speed of light. The much bigger issue is that most long distance paths on terrestrial networks go through a lot of routers and switches and other networking equipment, and each of these "network middleboxes" adds processing delay. A single overloaded router in your path could easily add 20ms itself, which is as much delay as four thousand miles of signal travel at the speed of light.

Starlink is more of an outlier, because nothing is "close" by terrestrial standards (being 300-400 miles up).

Starlink's satellites are indeed 300 to maybe even 800 miles up, but it's just 1 hop. So it takes 2-4ms to get from your home Starlink terminal on the ground to the satellite, and other 2-4ms to get from the satellite to Starlink's ground station, and the rest of the quoted "20-40ms" figure must come from processing delay in the one satellite itself, and in Starlink's terrestrial network.

Mentioning this number helps analysts realize that a LEO constellation at 300-800 miles is WAY better than old geostationary satellite broadband like HughesNet, for which the satellites are 22 thousand miles away, and where, after processing delays and other sources of latency, minimum ping times are in the high hundreds of milliseconds, to even a full second(!).

I know it's idealistic, but a far more useful number would be something like "% latency added", where it measured latency differences between theoretical maximum speed (unimpeded light travel) and observed latency. This would inform the consumer how well/poorly their specific network/ISP is performing in a general sense

This is basically what they're doing, once you realize that the latency added is a constant, not a percentage. That is, Starlink always adds 20-40 ms, no matter whether the rest of your path latency is tiny or huge. This is at least true for now, since the satellite-to-satellite global mesh stuff isn't operational yet. Starlink's latency-add doesn't scale with the overall latency of whatever path across the Internet your packets need to take, so it wouldn't make sense to give as a percentage. In comparison, as a Comcast Xfinity customer, when I ping their nearest router (presumably their CMTS) from my cable modem, I get a 10ms RTT, so DOCSIS as a customer-edge broadband access network technology adds 10ms in my case. So I can take that 10ms latency of DOCSIS as a residential broadband access technology and compare it to the 20-40ms latency of Starlink as a residential broadband access technology, and come to an understanding that if I switched from Xfinity to Starlink, I would expect my Internet ping times in general to have a net increase of 10-30ms.

Answer 2

Ping is indeed the "round trip time" for a connection, and for most modern landline connections it is so minimal that it makes nearly no difference. I remember a time of dial-up modems when pings of 200 to 300ms was "good".

The problem is that for most tasks such as watching video or web browsing "ping" is irrelevant, but for time critical tasks like games it can literally be the difference between life and death.

Take the scenario of a group of low latency players all playing a first person shooter together. They all have similar "ping" to the server, so they all get updates around the same time, and their "requests" for movement all go back quickly as well. Everything is fair and everyone has fun.

Now imagine a service that always adds 200ms to a players ping,and it is only that player because they have this particular service. That player will always be at least 200ms behind what every other player sees, and their responses will be slowed down as well due to working on already out of date information on what is going on. 200ms is huge. For professional players even 40-60ms can be big. By the time you've seen them they could have spotted you and fired off a few good shots and killed you. You end up screaming "how could they even see me?!" and they just go "you were just standing there."

At 60Hz your graphics card renders a new image every 16ms. A gamer playing at 120Hz will be seeing new images every 8ms, and adding 60ms from the ISP will be several frames behind what is actually happening.

20-40ms is probably not too bad for an ISP, but certain people will want to know "how bad" it could be.

They are simply giving you a "final leg" time in an explicit fashion because most other connections do not need to state it. Most landline connection between the home and local street cabinet have latencies measured in single digits, starlink may well be far better than traditional satellite services, but it will still be far worse than landline connections.

The reason it is not a percentage is because that "final leg" of 20-40ms is a fixed window. You might have an additional latency of 20ms once outside your carrier network, or you might have an additional 600ms to get to the dark seedy internet on the other side of the planet, neither of those destinations will change your initial journey of 20-40ms from your local ISP connection, and it isn't a simple percentage difference.

The 20-40 ms is a constant addition on top of the existing infrastructure.

Answer 3

Simply, it's round-trip latency to the first hop beyond the customer's link to the service provider, which sets a floor on the latency for anything else. This number is in the ballpark of:

• <1ms for a top-tier ethernet or fiber connection
• <10ms for good residential broadband
• 10-30ms for the lower tier of residential broadband
• 20-50ms for decent 4G wireless
• ~200ms for dialup or 3G
• >500ms for "traditional" (geosynchronous) two-way satellite internet

Of course the latency to anywhere on the internet can be more than this due to additional hops taken and distance covered from the provider's POP to the destination (anywhere from single-digit milliseconds on up to hundreds of milliseconds), but it can't be less. So, Starlink is trying to characterize their system, latency-wise, as being comparable to 4G, better than dialup or 3G, and an order of magnitude better than satellite systems that some of their potential customers might already have.

300-400 miles is quite a short distance from a speed-of-light perspective: less than 2 ms. Particularly, it's very much shorter than the 22,000 miles or so to geosynchronous orbit (which is about a 120ms trip each way). A round-trip request has to take four such hops (up to the satellite, back down to the provider ground station, out to somewhere on the internet and back, up to the satellite, and back down to the customer), which is the real killer for data via geosync satellite.