Running a Tor Relay or Bridge on a Romania VPS: Bandwidth, Abuse and What Hosts Allow

Short answer: a non-exit relay or an obfs4 bridge runs fine on a modest Romania VPS — it is mostly a bandwidth and uptime question, not a CPU one. An exit relay is a completely different decision: it attracts abuse complaints, gets the IP listed in blocklists, and needs explicit permission from your provider before you deploy it. If you want to contribute capacity to the Tor network from a cheap, privately-paid VPS, start with a middle relay or a bridge, and never run it on the same server as anything else you care about.

This article covers the practical side: how much relayed traffic your bandwidth quota actually buys, how to cap it in torrc so you don't blow through the allowance, what changes legally and operationally when you add an exit policy, and what to confirm with a host before you pay.

Relay, bridge, or exit: three very different jobs

Tor volunteers run three broad roles, and the hosting requirements differ sharply.

Guard/middle relay. Carries encrypted traffic between Tor clients and other relays. Nothing ever leaves your IP toward the open internet, so your server never appears as the source of anyone's browsing. This is the lowest-risk way to contribute. Your IP is published in the public consensus — the signed list of all relays that every Tor client downloads — so it's not a secret that you run one.

Bridge. An unlisted entry point for users in countries that block the public relay list. With the obfs4 pluggable transport, traffic is disguised so it doesn't look like Tor on the wire. Bridges typically use far less bandwidth than a busy relay, don't appear in the public consensus, and carry the same near-zero abuse risk as a middle relay. For a 1 GB VPS this is often the best fit.

Exit relay. The last hop that actually makes connections to destination sites on behalf of Tor users. Your IP is what the destination sees. That means complaints, scraping and brute-force reports, blocklisting, and occasional law-enforcement inquiries routed to whoever owns the IP. Most hosts — offshore or not — either forbid exits outright or require you to ask first and run them on dedicated ranges.

Bandwidth is the real constraint

Tor relays are measured in sustained throughput, but VPS plans are sold in monthly transfer quotas. Converting between them is where most new operators get surprised.

One megabit per second, sustained in one direction, is about 320 GB per month. A relay reads and writes roughly the same volume, so 1 Mbit/s of relayed traffic consumes about 650 GB/month across both directions.

Against flat monthly quotas, that works out to roughly:

• 1 TB — about 3 Mbit/s if only outbound is metered, ~1.5 Mbit/s if both directions count. Comfortable for an obfs4 bridge; a very small relay.
• 4 TB — about 12 Mbit/s outbound-metered, ~6 Mbit/s if both count. A respectable middle relay.
• 8 TB — roughly 24 / 12 Mbit/s on the same math.
• 16 TB — roughly 49 / 25 Mbit/s.

Two things to settle before you commit. First, ask the provider whether the quota counts outbound only or both directions, because that single answer halves or doubles your usable relay speed. Second, ask what happens at the limit — hard cutoff, throttling, or overage billing. A relay that keeps running past a metered cap is the fastest way to turn a €9.99 server into an unpleasant invoice. If you want to work through the numbers for your own case, our breakdown of how far 1 TB to 16 TB of VPS bandwidth actually goes has the full arithmetic.

Capping Tor to your quota

Tor has built-in accounting, and you should use it. In torrc:

RelayBandwidthRate 2 MBytes
RelayBandwidthBurst 4 MBytes
AccountingMax 3500 GBytes
AccountingRule sum
AccountingStart month 1 00:00

RelayBandwidthRate is the long-term average ceiling; RelayBandwidthBurst allows short spikes. AccountingMax with AccountingRule sum counts read plus write together against one number — which is what you want when your host meters both directions. Without sum, Tor applies the limit to reads and writes separately, effectively letting you use twice what you intended. Leave real headroom below the plan quota (the example assumes a 4 TB plan) for package updates, backups and monitoring.

When the budget runs out, Tor hibernates: it stops relaying, tells the directory authorities it's going offline, and wakes at the start of the next accounting period. That's a clean failure mode — much better than a surprise suspension.

CPU, RAM and IP requirements

Tor's relay workload is largely single-threaded for the cryptographic handling of each connection, so single-core performance matters more than core count. At the throughput levels a metered VPS supports — single-digit to low-double-digit megabits — a single modern vCPU is not the bottleneck.

RAM is modest but not trivial. A non-exit relay handling moderate traffic is generally comfortable in 512 MB–1 GB; busier guard relays holding many simultaneous connections want more. A bridge is lighter still. If you plan to squeeze a relay onto a 1 GB server alongside anything else, read our notes on tuning low-RAM VPS instances first — the OOM killer reaping tor at 3 a.m. is a silent uptime killer, and uptime is what earns relay flags.

You also need a dedicated, routable IPv4 address with a stable ORPort — relays behind shared NAT don't work. IPv6 is a bonus: with a /64 block you can advertise an IPv6 ORPort as well, which the network genuinely needs.

Abuse risk: non-exit vs exit, honestly

For a middle relay or bridge, abuse reports are rare to nonexistent, because no third party ever sees your IP as the origin of a connection. The practical risks are mundane: bandwidth overruns and the fact that your IP is publicly identifiable as a Tor relay.

Exit relays are where it gets real. Expect automated complaints about scanning, spam attempts and copyright-style notices, all pointed at the IP's owner — which is your host, not you. Under EU law, a pure conduit generally isn't liable for traffic it merely transmits, and Romania applies EU notice-and-action rules rather than auto-nuking servers on first complaint. That is a meaningful difference from hosts that suspend on receipt, but it is not immunity, and it does not override your provider's own terms. Our guide to how abuse complaints and takedowns actually work on a Romania VPS walks through the process and timelines.

The honest rule: do not deploy an exit relay without asking your host in writing first. Any provider that hasn't explicitly said yes will treat it as a surprise, and surprises end badly.

Why a relay should never share a server with your other projects

The moment your relay appears in the consensus, its IP is on a public list that commercial services consume. Expect CAPTCHAs, outright blocks from CDNs and anti-fraud systems, and mail that lands in spam folders. If that same server also hosts your website, your VPN exit, or your backups, you've just degraded all of them.

There's also a correlation angle. A relay is loud by design — public IP, public nickname, public ContactInfo if you publish one. Anything else on that box inherits the attention. Relay operators are encouraged to publish contact details so directory authorities can reach them, but a contact address is a real identity surface; use a dedicated, compartmentalised one or none at all, accepting that an anonymous relay may get less trust from the network.

The clean pattern is a cheap, dedicated server per role. At €4.99/mo for Iron 1 (1 vCPU, 1 GB RAM, 15 GB NVMe, 1 TB transfer), running a bridge on its own instance costs less than lunch, and nothing else on your infrastructure shares its reputation. If you're weighing that split, see one VPS or two: separating VPN, hosting and backups.

Worth noting for this specific use case: IronBalkans accounts require no email, name, phone or ID — just a generated account ID and a recovery key — and billing is in Monero, Bitcoin or Litecoin, with deployment in under 60 seconds once payment confirms. Every plan includes a dedicated IPv4, a /64 IPv6 block and DDoS filtering, which is the minimum a relay needs. If you want an exit policy rather than a middle relay, ask over Telegram or SimpleX before you pay, and get a clear yes.

A minimal non-exit relay configuration