What "Good" Looks Like: Healthy Ranges for LAN, Wi-Fi, WAN, and Internet Paths
"Latency to the server is 9 milliseconds." Quick: is that good? The honest answer is that you cannot know without more context. If that server sits two racks away on your wired LAN, 9 ms is alarmingly bad — something is queuing or broken. If it sits in a data center three countries away, 9 ms would defy physics. The number is meaningless until you know what kind of path produced it.
That context is exactly what this guide provides. It is a reference for the four core metrics — latency, jitter, packet loss, and throughput. It covers the four kinds of path you actually manage: the wired LAN, business Wi-Fi, site-to-site WAN and VPN links, and general internet paths. You get one master table to bookmark, and then, more importantly, the reasoning behind every number. That means the physics, the medium, and the distance that make each range what it is. Memorized numbers expire; understood numbers let you judge situations the table never anticipated.
Two warnings before you bookmark anything. First, generic ranges are a sanity check, not a verdict — the baseline of your network beats any table, including this one. Second, a number measured carelessly will compare uselessly. Both caveats get their own sections at the end; please read them before you wave this page at a vendor.
In this guide
- The four numbers, in one minute
- The reference table
- Why a gigabit link never moves a gigabit of files
- Wired LAN: the gold standard
- Business Wi-Fi: physics on a party line
- Site-to-site WAN and VPN: distance rules
- Internet paths: someone else's network
- Your normal beats generic normal
- Measure properly before you compare
The four numbers, in one minute
Latency is travel time: how long a packet takes to reach its destination and for the reply to come back, measured in milliseconds (ms). Tools like ping report this round-trip time (RTT), and every range in this guide is a round trip unless it says otherwise. Jitter is how much that travel time varies from one packet to the next. A path with a steady 40 ms is healthier for calls than one swinging between 10 and 70, even though the averages match. Packet loss is the percentage of packets that never arrive and is the harshest of the four: small numbers hurt. Bandwidth is the capacity of a link; throughput is what you actually achieve through it. The gap between the two is protocol overhead, which gets its own section below. Each metric has full guides elsewhere in this library; that one paragraph is enough to read the table.
The reference table
These are sustained values on a healthy path under normal load, measured properly over time (see the final section). Brief blips beyond them are ordinary life on a packet network; readings that sit outside them are telling you something.
| Environment | Round-trip latency | Jitter | Sustained loss | Realistic throughput |
|---|---|---|---|---|
| Wired LAN (same site) | Under 1–2 ms | Under 1 ms | Effectively 0% — treat any steady loss as a fault | 94–95% of link speed |
| Business Wi-Fi | 2–10 ms to the gateway, with brief spikes | 2–10 ms, spiky under load | Under 1% (after the radio's own retries) | Roughly 30–60% of the advertised rate, shared per channel |
| Site-to-site WAN / VPN | Distance-driven: ~1 ms per 100 km of route, plus a few ms for equipment | Under 10 ms | Under 0.1%, ideally zero | 80–95% of the circuit after VPN overhead |
| Internet paths | 10–40 ms regional; 20–80 ms same continent; 80–250 ms intercontinental | Under 30 ms | Under 0.5% sustained; near zero most of the time | 85–95% of the subscribed rate, to nearby servers |
The rest of this guide explains where each row comes from and what to do first when a reading falls outside it.
Why a gigabit link never moves a gigabit of files
Start with the throughput column, because it confuses everyone once. Every chunk of user data travels inside nested envelopes. The application's data goes inside a TCP segment, which goes inside an IP packet. That goes inside an Ethernet frame, which is sent with a preamble and followed by a mandatory gap before the next frame. Every envelope costs bytes, and those bytes spend the link's capacity without carrying any of your file:
one full-size packet on gigabit Ethernet: user data (TCP payload) 1,460 bytes + TCP header 20 bytes + IP header 20 bytes + Ethernet header and checksum 18 bytes + preamble and inter-frame gap 20 bytes -------------------------------------------------- total spent on the wire 1,538 bytes efficiency: 1,460 / 1,538 = ~95% 1 Gbit/s link => ~940-950 Mbit/s of data = ~117 MB/s in a file copy
So on a perfect wired path, about 94–95% of the sticker rate is the ceiling — before any other traffic, retransmissions, or distance effects. Two follow-on facts save you from common panics. First, links are sold in megabits per second but file managers report megabytes, which differ by a factor of eight. A gigabit link showing "117 MB/s" is not losing 88% of its speed. It is behaving perfectly. Second, every extra envelope costs more. A VPN wraps each packet in yet another header (dropping efficiency by several percent more). Wi-Fi loses far more again — not to headers but to airtime rules, as its section explains.
Wired LAN: the gold standard
Expected: round trips under 1–2 ms anywhere in the building, jitter under a millisecond, loss indistinguishable from zero, and single-transfer throughput at 94–95% of the port speed.
Why these numbers: distance is simply not a factor indoors. Signals cross even a large building in a few millionths of a second, and a healthy switch forwards a frame in microseconds as well. So nearly all of that 1–2 ms budget is software overhead at the endpoints, not travel. Modern wired Ethernet is also full duplex. Each cable carries traffic in both directions at once over a dedicated path to the switch, so nothing competes for a turn to talk. And good cabling flips bits so rarely that corrupted frames are a rounding error. Put together, there is no innocent explanation for LAN readings like 30 ms or 0.5% loss. Wired LANs do not have "a bit of loss" the way roads have a bit of traffic. Steady loss on a LAN means a specific thing is broken or a specific queue is full.
If a number is outside range: check the path's port error counters first. A rising count points at a bad cable, a failing port or optic, or a speed/duplex mismatch. Next, check utilization on the uplinks the path crosses; a saturated uplink adds queuing delay that innocent LAN hardware never would. If the counters and uplinks are clean, suspect the endpoint itself. An overloaded server answers pings late, and that is a host problem wearing a network costume.
Business Wi-Fi: physics on a party line
Expected: expect 2–10 ms to the gateway most of the time, with occasional spikes into the tens. Expect jitter of a few milliseconds that spikes under load, and loss under 1% as applications see it. Expect real throughput of roughly a third to two-thirds of the advertised rate — shared among everyone on the channel. Half the sticker is a fair planning number.
Why these numbers: Wi-Fi is a party line. The radio channel is half duplex and shared: only one device can transmit on it at a time, including the access point itself. Every device must listen for a quiet moment, wait a small randomized delay to avoid talking over others, then transmit. And that queue-for-the-microphone dance is latency and jitter built into the medium, present even on a healthy network. Radio frames are also lost routinely (interference, distance, a colleague's misbehaving gadget), so Wi-Fi acknowledges every frame and silently retries the failures. Retries convert most radio loss into delay before your applications ever see it. That is why bad Wi-Fi usually shows up as jitter first and loss only when things get truly ugly. Throughput expectations need the same discount: advertised rates are best-case signaling rates at short range on a clean channel. Real distance and walls lower each client's rate. Slower or farther clients spend more airtime per byte — the whole room slows toward its slowest talker.
If a number is outside range: establish scope first: one client, or everyone on that access point? For one client, look at its signal strength and negotiated rate — distance and dead spots explain most single-victim misery. For everyone, check the AP's client count and channel utilization, and look for a neighboring AP (yours or a neighbor's) camped on the same channel. Then repeat the measurement from a wired port near the AP: if the problem survives the cable, stop blaming the air.
Site-to-site WAN and VPN: distance rules
Expected: latency is set almost entirely by distance. Expect roughly 1 ms of round trip per 100 km of cable route, plus a few milliseconds for equipment and the VPN itself. Nearby cities land in the single digits to low tens; cross-country links in the tens. Jitter under 10 ms on a healthy link, sustained loss under 0.1% and ideally zero, and throughput at 80–95% of the circuit once VPN overhead is paid.
Why these numbers: light in fiber covers about 200,000 km per second — two-thirds of its speed in a vacuum. That works out to the 1 ms-per-100 km round-trip rule. That number is a law, not a vendor setting: no upgrade shortens it. Real routes also run longer than the map suggests, following roads, railways, and seabeds rather than flying crow-straight. So a link between cities 300 km apart may ride 500 km of glass. On top of physics, each router hop adds a little processing. A VPN adds two costs of its own. One is encryption work on the endpoints (CPU time per packet). The other is one more envelope around every packet (bytes, and occasionally awkward packet-size issues on the tunnel). A dedicated business circuit that is not congested should be boringly steady — low jitter, essentially no loss. So sustained loss or swinging latency on a WAN is a fault or a full queue somewhere, never "just how WANs are."
If a number is outside range: check utilization at both ends first. The most common "WAN problem" is one site's own uplink queuing, and it perfectly impersonates a slow circuit. Next, check the VPN endpoints' CPU during the affected hours; an overworked firewall encrypting at its limit adds delay that looks like distance. Then compare the current route against your baseline with a traceroute. A step change in latency with a new path in the middle means your traffic has been rerouted. That is the carrier's story to explain. Open that ticket with timestamps and data attached.
Internet paths: someone else's network
Expected: 10–40 ms round trip to regional destinations, 20–80 ms across a continent, and 80–250 ms between continents. The far end of each range belongs to the longest routes. Jitter under 30 ms is acceptable for general use. Loss should be near zero most of the time; brief blips happen, but anything sustained above roughly half a percent is worth investigating. Downloads from nearby, well-connected servers should reach 85–95% of your subscribed rate.
Why these numbers: an internet path is a relay race across networks owned by strangers. Your ISP, a carrier or two, and the destination's provider hand packets off at exchange points. Distance sets the floor (the same fiber physics as the WAN section), the handoffs add hops and variability, and congestion at busy hours sets the weather. You rent a path, not the path: routes change without notice, and each change can move your latency to a new normal. Loss on the internet is usually transient — a blip during a route change, a briefly congested exchange. That is why the tolerance here is looser than on links you own. Sustained loss to everywhere, though, is not internet weather; it lives in the one segment all your traffic shares, your own uplink or your ISP's edge.
If a number is outside range: scope before anything else: one destination or all of them, one user or the whole site? All-destinations problems start at home. Check your own uplink's utilization before suspecting the wider internet. A full local link is the most common cause of "the internet is slow" ever recorded. For one destination, test two or three unrelated targets to confirm the contrast, then traceroute toward the problem to see where the delay begins.
Warning: read traceroute with suspicion. Routers answer probes aimed at them with their spare capacity, as their lowest priority. So a scary spike at hop 4 that vanishes by hop 7 usually means hop 4 is lazy about replying, not slow at forwarding. Only a rise that starts at some hop and persists to the destination marks a real problem segment.
Your normal beats generic normal
Every range above was written without meeting your network. Your paths have personalities this page cannot know. There might be a VPN that rides a longer route than geography suggests. Or a Wi-Fi floor might have a microwave oven problem, or an ISP might reroute every night. Suppose your site-to-site VPN has sat at a steady 62 ms for as long as you have measured it. Is that "bad" because a table said your region should be under 40? No — it is your route's honest length, and 62 is your normal. The day that same link reads 95 ms at the same utilization, something real has changed — even though plenty of generic tables would bless 95 without comment.
This is baselining: let your monitoring record each important path for a couple of typical weeks. That way, every one has a known normal and a known daily rhythm. From then on, alert on change from baseline — latency 50% above its usual for fifteen minutes, loss where there was none — rather than on generic absolutes alone. Use this page's table for day one, before a baseline exists, and afterward as a gut check on the baseline itself. If your LAN's "normal" is 40 ms, you have not discovered a special network. You have documented a longstanding problem.
Measure properly before you compare
A reference range can only be as good as the measurement you hold against it. Before comparing, make sure the number deserves the trust:
- Measure over time, not once. A single ping burst is a weather report for one minute. Judge a path on hours or days of samples, spanning both busy and idle periods. Note which is which, because a path that is clean at midnight and miserable at 10 AM has told you exactly where to look.
- Give loss enough samples to exist. Loss is a percentage, so your sample count is your microscope's magnification. Twenty pings cannot see 1% loss at all; even 100 only barely resolve it.
- Take Wi-Fi out of the experiment. Unless you are measuring the Wi-Fi itself, test from a wired machine. Otherwise every number carries the air's fingerprints, and you will chase radio ghosts across your WAN.
- Measure the path the complaint is about. The monitoring server's path to a branch is not the users' path to their app. Measure from where the pain is felt to where the service lives, or as close to both as you can get.
- Mind bits versus bytes. An eight-fold "discrepancy" between a link's rating and a file copy's speed is unit conversion, not an outage. Convert before comparing anything to the throughput column.
- Treat speed tests as blunt instruments. A speed test measures the path to that one test server, while deliberately flooding your link. This is a self-inflicted congestion event that distorts every other metric while it runs. Fine for a rough capacity check; useless as evidence of loss or jitter.
Tip: a quick rule for loss measurements: to trust a reading of X percent, send at least a few hundred packets. To see 0.1% at all you need a thousand. When a monitoring system reports 0.2% loss from its regular small samples, read that as "some loss exists." In that case, run a longer test before quoting a figure to anyone.
Where to go next
Bookmark the table, but remember what it is: the start of the argument, not the end. The ranges tell you whether a number deserves attention. The reasons behind them tell you where to look next. Your own baseline outranks both. From here, the deep dives on Network Latency and Packet Loss cover measuring and fixing the two troublemakers in detail. The rest of The Core Network Metrics explains how the four numbers interact and which one matters for each application you run. The full library lives at the topics index.
Put this into practice with SysPeep Monitor
Reference ranges get you through day one; your own baseline is the real yardstick. SysPeep Monitor records response times, packet loss, and interface utilization around the clock from one central Windows console. That way, every path on your network develops a known normal. SysPeep's thresholds and alert rules tell you the moment a number drifts away from yours. The free trial takes minutes to set up.
Download Free TrialLearn About SysPeep