Side by side, the real difference is
who controls the control plane

This page only states things that can be checked: the limits and prices published on each vendor's own pricing page, public architecture, and what "the control plane sits in someone else's data center" actually costs you. Figures captured on 17 November 2026. Found an error? Email us and we will verify and correct it.

Five approachesVendor-published numbersNo invented benchmarksWeaknesses included
Start here

Which row looks like you?

Five situations cover most of the decisions. If one of them is yours, you probably do not need to read further.

Your situationBetter fitWhy
Topology data must not leave your own data centerVirtuaMesh / HeadscaleBoth run the control plane on your server; the difference is how much product you get out of the box
5–50 people, need role-based access and audit logs, no dedicated ops teamVirtuaMesh StandardSelf-hosted control plane plus a GUI, RBAC, tiered ACL and audit logs
Just want a few free devices and do not care where the control plane isZeroTier (10 devices) or Tailscale (unlimited devices, non-commercial)Both free tiers are more generous than ours. That is a fact, not a sales angle.
Procurement process, local invoice and vendor SLA requiredPgy (Oray)Hosted in China with multi-line BGP relay, priced per licensed device per year
You can operate a server and only want a controllable WireGuard control planeHeadscaleFree and open source, no user cap; you pay in operations time
Not sure you need a mesh at all?

If the real question is "should we do this, and in what order", start with the buying guide — it subtracts with three questions, then you come back here to pick a product.

To be clear: free-tier generosity is not our advantage.

ZeroTier gives 10 devices free; Tailscale's free plan has unlimited devices. Ours is 5 nodes. If device count for zero cost is your only criterion, pick one of those. We fit the case where you need to own the control plane without building one from scratch.

How to evaluate

Comparing price and latency alone usually means redoing it in six months

The differences that decide long-term cost are not in the spec sheet.

01 · Where the control plane sits

Who decides whether your devices connect

The control plane holds identity, keys and policy. If it lives in a vendor's data center, their account policy, pricing changes or service retirement reach into your network. If it lives in yours, that risk becomes yours to operate.

02 · After hole punching fails

Which machine your traffic falls back to

Direct peer connections are fast everywhere. The gap appears on fallback: a relay overseas turns tens of milliseconds into hundreds, while a relay in your own rack is only as slow as your bandwidth.

03 · What you are billed for

Devices, seats, or licensed units

Per-device pricing favours many machines and few people; per-seat pricing favours the opposite. The same requirement can cost twice as much depending on the model.

At a glance

How the five options actually differ

Limits and prices come from each vendor's pricing page, captured 17 Nov 2026. "Entry price" is the cheapest self-serve tier listed, excluding hardware and add-ons.

OptionControl planeFree tier (vendor published)Entry priceRelay on fallbackAudit & roles
TailscaleVendor-hosted6 users, unlimited user devices, 3 ACL groups, 50 tagged resources; non-commercial only$8 / user / monthVendor DERP (self-hosted peer relay optional)ACL groups by tier; flow logs on Premium
ZeroTierVendor-hosted10 devices, 1 network, 1 network admin; personal non-commercial$18 / month (10 devices, then $2 each)Vendor root serversAudit logs on paid tiers (from 30 days)
Pgy (Oray)Vendor-hosted in ChinaNo free tier listed on pricing page¥148 / year / licensed deviceVendor BGP relay linesLog retention by tier: 30 / 90 / 180 days
HeadscaleYoursSoftware is free; you pay for the server~¥50–200 / month serverYour own DERPNo official UI; config files plus community panels
VirtuaMeshYours5 nodes, 2 networks, basic ACL¥399 / year (15 nodes, 3 networks, 1 exit node)
¥993 on a 3-year contract
Your own DERP clusterRBAC, four ACL tiers, audit retention by tier
One by one

What each option is good at, and what it costs you

Every paragraph below lists both. A comparison that only lists the first half is the most common kind of dishonest comparison page.

Tailscale

The best-polished clients, with the control plane hosted abroad

Good at: a WireGuard-based mesh with carefully built clients on every platform, MagicDNS, and identity through Google, Microsoft or GitHub, plus strong SSO and ACL expressiveness.

You accept: the free Personal tier is non-commercial only, so a company network must pay; billing is per seat, so headcount drives cost; and when hole punching fails, traffic falls back to vendor DERP relays, whose latency and throughput are outside your control.

ZeroTier

The most generous free tier, on a non-WireGuard protocol

Good at: 10 free devices, the friendliest of the group for many machines and few people; a layer-2 style network that suits broadcast and discovery protocols; mature clients including NAS and router firmware.

You accept: its own protocol rather than WireGuard, so performance and audit are separate from the WireGuard ecosystem; the free tier allows one network and one admin, which teams outgrow quickly; and a self-hosted controller means giving up vendor-side management and support.

Pgy (Oray)

The easiest domestic procurement, with data on the vendor side

Good at: multi-line BGP relay inside China, stable on weak networks; hardware routers and boxes, which suits stores that have no server at all; complete invoicing and support processes, so there is someone to call.

You accept: no free tier is listed on the pricing page, with the entry plan at ¥148 per licensed device per year; the control plane belongs to the vendor, so topology and policy are not yours; audit retention depends on tier (30 / 90 / 180 days).

Headscale

Free and fully yours, if you are the operations team

Good at: an open-source implementation of the Tailscale control server that works with the official clients, with no user cap. The cost is one server.

You accept: no official web console, so day-to-day work is command line and config files; upgrades, backups and incident response are yours; and its feature scope is narrower than the hosted product.

VirtuaMesh

Self-hosted and productised, but you need a server

Good at: control plane and DERP both run on your own server, so topology and policy never leave your rack; a graphical console with RBAC, tiered ACL and audit logs built in rather than assembled; Chinese-language UI and domestic network conditions as a design premise.

What it will not do: it needs a server, so "zero deployment" users are not a fit; the free tier is 5 nodes, fewer than ZeroTier's 10 devices; client polish and third-party integrations are behind Tailscale; and we publish no benchmark numbers — the next section tells you how to measure it yourself, which is worth more than a millisecond figure of unknown origin.

Price

Why per-unit prices cannot be compared directly

A Pgy licence includes their relay servers and bandwidth. Ours does not include your server. Put those two numbers side by side and the conclusion will be wrong. So here is a three-year total, with both deployment paths listed separately.

Three-year costPgy Commercial
(5 licensed devices)
VirtuaMesh Standard
(spare machine)
VirtuaMesh Standard
(cloud host)
Nodes / licences5 client licences (= 5 devices; hardware licence ¥328/yr each)15 nodes15 nodes
Licence / subscription¥148 × 5 × 3 = ¥2,2203-year contract ¥993
(¥399/yr × 3, 8.3折)
same, ¥993
ServerNot required, included¥0 (existing machine)≈ ¥600/yr × 3 = ¥1,800
Relay bandwidth8Mbps per licensed device, included; expandableDepends on your machine and lineCloud host bandwidth
OperationsVendorSelf-service (about half a day to deploy)Self-service (about half a day to deploy)
Three-year total¥2,220¥993¥2,793
Half the price without labour, the same ballpark once you add it

The figures above leave no allowance for labour or extra bandwidth. At ¥200/hour (half a day to deploy plus two hours a year), self-hosting adds roughly ¥1,400–2,000 over three years: about ¥2,400–3,000 on a spare machine, ¥4,200–4,800 on a cloud host.

Per node the gap is clearer: Pgy works out at ¥2,220 / 5 ≈ ¥444 per device over three years (including their relay bandwidth and lines); ours is ¥993 / 15 ≈ ¥66 per node (excluding your server and operations). The larger the scale, the wider the gap.

In other words: how much you save depends on whether you already have a machine and someone who can look after it. With a spare machine and in-house hands, three years costs about half of Pgy's total; with only a cloud host and labour costed in, the two land in the same range (¥2,400–3,000 against ¥2,220). We would not lead with "cheaper" — what you actually buy is the next section.

Prices captured from vendor pricing pages on 17 Nov 2026. Pgy is costed at its published annual price, without any multi-year discount it may offer; VirtuaMesh uses our published 3-year contract price (8.3 of 10). Cloud host estimated at a 2 vCPU / 2GB entry configuration, bandwidth costed at 5Mbps. The labour estimate is illustrative only — your hourly rate sets the real number.

Bandwidth and data

How fast a relayed path runs depends on where the relay is

This is where the options differ most, and where "unlimited traffic" quietly hides a bandwidth cap. Traffic may be unlimited while throughput is not.

OptionWhere relay bandwidth comes fromIs there a bandwidth commitmentWhere topology and logs live
Pgy (Oray)Vendor BGP relay lines, allocated per licensed deviceYes: 8Mbps per device on Commercial, 15 on Flagship, 20 on Platinum, 35 on High-performance; paid expansionVendor side; audit logs retained 30 / 90 / 180 days by tier
TailscaleShared vendor DERP; self-hosted peer relay uses your own bandwidthNo per-user throughput commitment; best effortVendor control plane; flow logs on Premium
ZeroTierVendor root-server relaysNo published per-user bandwidth figure; lower tiers are capacity-limitedVendor control plane; audit logs on paid tiers
HeadscaleYour own DERP deploymentCeiling = your server and lineEntirely in your network
VirtuaMeshYour own DERP clusterCeiling = your server and line (plannable, expandable)Entirely in your network
This is not entirely in our favour, so it is worth stating plainly

If the business genuinely needs "a steady 20Mbps even when hole punching fails", a contractual bandwidth commitment like Pgy's is less effort — what you buy is somebody else's BGP lines. Self-hosted bandwidth is your own: a fat link is fast, a thin one is slow, and relay traffic competes with business traffic on the same line. Before choosing it, check that the machine bandwidth and monthly allowance are actually enough.

Control

What owning the control plane actually buys you

"Self-hosted" can sound like a slogan. It maps to four verifiable outcomes, and they are the only core difference we can defend.

01 · Data stays inside

Who is online, who is connected — never through a third party

Device inventory, keys, access policy and connection records live in your own database. There is no endpoint for a vendor to look at them — not a promise not to look, but a path that does not exist.

02 · Supplier risk isolated

Price changes, billing changes, shutdowns do not reach you

With a hosted service, their account system, pricing policy and service lifecycle directly affect your network availability. Self-hosting moves that risk out of your dependency chain.

03 · Audit on your terms

Retention and access are your decision

Hosted audit capability is tied to a tier (30 days here, flow logs on a premium plan there). Self-hosted retention is limited by your disk, and exporting to a SIEM needs nobody's permission.

04 · Portable

No export request to raise

Configuration, device inventory and audit records are in your own database. Compliance reviews, internal audits or a future migration start from material you already hold.

The cost is inside those four points too

If the control plane is yours, upgrades, backups and incident response are yours. We wrap that in a GUI and installers, but somebody still has to be able to log in when it breaks. That is the inherent cost of self-hosting, not something a product can remove.

Performance

Do not trust any millisecond figure, including ours

Latency depends on both broadband lines, the NAT type at each end, and where the relay sits. The same product can differ tenfold between two networks, so an "average latency" claim only holds for the environment it was measured in.

Step 1

Confirm direct or relayed

Check the path shown in the client (VirtuaMesh shows direct or relayed). If both ends are domestic on the same ISP and it still says relayed, hole punching failed and the number you measure is not the product's ceiling. On the command line, traceroute shows whether the first hop is the peer's virtual IP.

Step 2

Measure latency, jitter and throughput

ping -c 100 gives average and mdev for latency and jitter; iperf3 -t 30 in both directions gives throughput. Testing download only overstates the experience — uploads and remote desktops depend on the uplink.

Step 3

Repeat during a real peak hour

Run it once during business hours and once between 20:00 and 23:00. Home broadband uplink typically drops at peak, and that variation is often larger than the difference between products.

Why there are no numbers on this page

Because we have not run a publishable, reproducible sample in our own environment. When we do, the commands, the environment and the results go out together — not a flattering figure first. What you measure with the three steps above is worth more than anything we could write here.

FAQ

Eight questions asked before deciding

Does self-hosting mean a dedicated ops person?

No. Deployment takes about half a day; after that it is upgrades and backups. Below a few dozen devices, most teams spend an hour or two per quarter. The trade-off is that someone must be able to log in when something breaks, instead of calling support.

Do users need training?

Sign in, approve the device, connect. The flow matches other tools in this category. Most staff install once, sign in once, and reconnect automatically afterwards.

Is a relayed connection always slow?

No — it depends where the relay is. A relay in your own rack on adequate bandwidth can feel close to direct; a relay overseas is where hundreds of milliseconds and tens of KB/s come from.

Can a contractor reach only one system?

Yes. ACL rules match source and destination subnets, protocol and port, so you open exactly one target. Disable the account when the project ends; the audit trail stays.

Migrating from Tailscale — do clients need reinstalling?

Yes. The control planes and key material are separate, so clients re-register against your VirtuaMesh control plane. Subnet plans carry over; devices rejoin.

What separates the free and paid plans?

Node count, network count and ACL tier. The free plan (5 nodes, 2 networks, basic ACL) is enough for a real pilot; rule limits, audit retention and exit nodes scale by tier.

Which clients are supported?

Windows, macOS, Linux, Android, iOS and OpenWrt. The server deploys with Docker Compose or a guided installer script.

Does our traffic really avoid your servers?

With a self-hosted deployment, the control plane is on your server and business traffic is encrypted end to end between devices. Relays forward ciphertext only: we can see neither your topology nor your traffic.

Verify it on your own network, with one device

You do not need to finish the comparison. Two devices, one open port, one real workflow will tell you more than ten reviews.

Sources: limits and prices quoted from the Tailscale, ZeroTier and Pgy (Oray) pricing pages, captured 17 November 2026. Headscale is an open-source project; cost estimated as a self-provided server. Vendor terms change; always check the current pricing page. If you find a factual error, email sales@virtuamesh.com and we will verify, correct and note the revision date.