Why v2

Protocol v0/v1 sent the server's full ref advertisement at the start of every fetch — a payload proportional to the number of refs. On repos with tens of thousands of refs (releases, PRs, mirrors), this dominated transfer time. Protocol v2 (default in Git 2.26+) lets the client request specific refs by name, dropping the upfront cost dramatically.

Verifying it is on

GIT_TRACE_PACKET=1 git ls-remote origin 2>&1 | head -10
git config protocol.version 2
git config --global protocol.version 2

Look for version=2 in the trace. Some older servers still negotiate down to v0 or v1.

Server requirements

• HTTP: Git 2.18+ on the server, with http.receivepack aware of v2.
• SSH: Git 2.18+ on both ends; for many providers, this is automatic.
• Some legacy hosts may need uploadpack.allowFilter=true and explicit v2 enablement.

Capabilities

v2 introduces capability negotiation. The client lists what it understands; the server responds with the intersection. Newer features (filter, ref-prefix, sideband) ride along.

GIT_TRACE_PACKET=1 git fetch 2>&1 | grep -i 'capabilit\|fetch='

Practical impact

On Linux's repo (~50k refs), a fresh fetch over v0 spent 4-8 seconds in advertisement; v2 cuts this to under a second. On JS monorepos with auto-generated PR refs, the difference is even more pronounced.

Sideband and progress

v2 standardizes sideband-64k progress reporting. Disable for non-tty consumers like CI:

git fetch --quiet
git -c progress.delay=999 fetch

Common mistakes

Forcing protocol version 0 in old config files left from upgrades. Believing v2 helps with object transfer size — it primarily helps with negotiation overhead. Mixing v2 client with v0 server: works (negotiates down) but wastes the win.

Combine with

• Negotiation algorithm: fetch.negotiationAlgorithm=skipping.
• Reachability bitmaps on the server.
• Ref filtering: fetch.refSpecs.

Related

See "Wire protocol optimization: stateless-connect" and "Fetch and push over large repositories".