中文

Reviews

SS vs VMess vs Trojan vs Hysteria2: which to use

What actually differs between the five common proxy protocols, where each one wins, and why you should keep two different kinds on hand rather than picking a favourite.

By MaoUpdated 2026-08-246 min read

Day to day, your client handles all of this. Two situations make it worth knowing anyway: when one protocol stops working, and when choosing a provider.

The whole thing in one table

ProtocolTransportDisguiseOverheadUnder packet lossWhere it stands
ShadowsocksTCPWeakLowestPoorVeteran; still good for low latency
VMessTCPMediumMediumPoorLegacy; works, not a first choice
VLESS + RealityTCPStrongestLowPoorThe current recommendation
TrojanTCPStrongLowPoorDependable, widely supported
Hysteria2UDP/QUICMediumMediumBestThe answer on bad connections
TUICUDP/QUICMediumMediumGoodSimilar to Hysteria2

One at a time

Shadowsocks — oldest, not obsolete

Strengths: lowest overhead, lowest latency, broadest client support, simplest configuration — four parameters and you are done.

Weakness: bare SS has a relatively distinctive signature and gets targeted on networks doing deep inspection.

Use it for:

VMess — the middle generation

Strengths: complete feature set, multiple transports (tcp/ws/grpc), mature ecosystem.

Weaknesses:

Use it for: whatever you already have. Vmess is still common among free nodes for historical reasons.

Always paired with ws + tls; bare vmess is a waste of time. See Free V2Ray nodes.

VLESS + Reality — the current best answer

What new deployments choose.

VLESS strips VMess's own encryption and leaves it to TLS, lowering overhead.

The real advance is Reality: instead of presenting its own certificate, it borrows a genuine website's TLS handshake. What an observer sees is indistinguishable from a visit to that real site.

Strengths: best disguise available, good performance. Weaknesses: more involved to configure; older clients may not support it.

Trojan — the low-drama option

How it works: presents as ordinary HTTPS on port 443. The handshake is close enough to a normal website visit that blocking it means damaging legitimate traffic.

Strengths: stable, near-universal client support, simple configuration. Weaknesses: needs a real domain and certificate; disguise is slightly behind Reality.

My view: the one that causes the fewest problems. Not the most impressive on paper, rarely the source of a bad evening.

Hysteria2 — when the line is bad

How it works: built on QUIC (UDP), with its own congestion control and forward error correction.

The problem it solves: TCP responds to loss with retransmission and congestion backoff, and throughput collapses. Hysteria2 absorbs a few dropped packets and keeps going.

Where it wins big:

SituationWhy
Residential peak hoursMany home lines hit 3–10% loss in the evening
Mobile networksMuch higher loss than fixed lines
Mediocre long-haul routesMore hops, more chances to drop

Measured on the same server during an 8% loss window: TCP-based protocols managed 3 Mbps, Hysteria2 held above 40.

Where it fails:

TUIC — same family

Also QUIC-based, similar reasoning. In practice close enough to Hysteria2 that you should just use whichever your provider offers.

By situation

Browsing and AI tools

Trojan or VLESS+Reality. Stability above everything; AI tools break badly on dropped connections.

And pin the node manually rather than letting a group auto-switch — reasoning in Using ChatGPT over a free VPN.

Video

Depends on your line:

Gaming

Latency and jitter, not bandwidth.

Make sure the node config has udp: true, or game traffic will not pass.

Routers and transparent proxies

Trojan or SS, because router CPUs are weak and overhead matters.

Hysteria2's encryption can outrun a low-powered device before it saturates the link.

Heavily inspected networks

VLESS + Reality. Best disguise currently available.

Keep two kinds

Have one TCP-based node and one UDP-based node available at all times.

They fail for different reasons:

My client holds:

Switching takes a second, which beats hunting for a replacement mid-task.

When choosing a provider, prefer one supporting several protocols — a single-protocol service is a single point of failure. More on that in Luobo Cloud after three months.

If you want one sentence

Check that your provider offers Hysteria2 and either VLESS or Trojan. Beyond that, let the client deal with it.

Questions people keep asking

Which protocol is fastest?

None of them, in general. On a healthy connection the differences are small and the bottleneck is server bandwidth. On a lossy connection, QUIC-based protocols like Hysteria2 pull ahead substantially, because forward error correction means a few dropped packets do not collapse throughput.

Which is hardest to detect?

VLESS with Reality, currently. It borrows a real website's TLS handshake, so there is almost nothing distinctive to fingerprint. Trojan is close behind, presenting as ordinary HTTPS. Bare Shadowsocks and VMess have more recognisable signatures.

Why will my Hysteria2 node not connect?

Usually because your network restricts UDP, which is what QUIC runs on. Some campus and corporate networks block it outright, and it is disrupted at certain times on others. Test by trying a TCP-based node from the same provider; if that works and Hysteria2 does not, UDP is your problem.

Do I need to understand this to use a subscription?

No. A subscription and a modern client handle it for you. Knowing a little helps in two situations: when one protocol stops working and you need to know what to switch to, and when choosing a provider and deciding what to look for.

Written from my own testing. Conditions change; if something here has gone stale, email [email protected] and I will update it with a date.

Read next