xrayGraphDB v5.0 property values, strings, WAL, keys: AES-256-GCM at rest every connection names its database licensed only
Graph engine · one server

All of Friendster, end to end: 65.6 million people, 22 hops deep.

xrayGraphDB holds and traverses the entire Friendster social graph, 1.8 billion friendships, on one server. A breadth-first walk from one person reaches every one of 65,608,366 people, and the last of them is 22 hops away. Every property value, string, WAL entry and key is encrypted at rest with AES-256-GCM.

People
65,608,366
Friendships
1,806,067,135
Directed edges
3,612,134,270
Dataset
com-Friendster (SNAP)
friendster · breadth-first from user 101 across the whole graph people first reached at that hop
15
hopnewly reached (log scale)peoplecoveredtime
Benchmarks

Numbers we can show you the raw samples for

A figure appears here only when its run archive recomputes it from the raw samples. Courtroom-clean latency and throughput results are coming soon, measured head-to-head on equal hardware. (Hint: we think you’ll like them.)

Scale · one server● verified
3.6Bdirected edges loaded and traversed
datasetcom-Friendster (SNAP), the LDBC Graphalytics copy
vertices65,608,366
edges1,806,067,135 undirected · 3,612,134,270 directed
Durability · per commit● by design
  1. 1Property values and the WAL entry sealed with AES-256-GCM, keys wrapped
  2. 2Write forced to disk with fsync
  3. 3Commit acknowledged to the client

Benchmarks run in this mode. There is no faster, unencrypted mode to quote.

Point lookup · p50 / p99
— ms
○ re-measuring on equal hardware
k-hop traversal · p50
— ms
○ re-measuring on equal hardware
Keyed-edge ingest · per batch
— ms
○ re-measuring on equal hardware
How a number gets onto this page
  1. step 1

    Protocol frozen

    Hashed and timestamped before the first sample.

  2. step 2

    Binary pinned

    The running engine is checked against the build under test.

  3. step 3

    Samples signed

    Raw samples go into a signed, hash-chained ledger.

  4. step 4

    Recomputed

    Every published figure is recomputed from those samples.

How it works

One trace across three public datasets

The Tracing demo joins federal campaign-finance filings, nonprofit tax returns and the ICIJ Offshore Leaks into one graph, then follows the connections from any starting point.

Each source arrives as its own bulk files. xrayGraphDB stores every filing as nodes and edges, records which file and snapshot each fact came from, and links entities across sources only where their filings share an exact identifier or address.

A trace starts from one entity and expands outward hop by hop. At every hop the engine reports how many entities it reached and how many connections it explored, and the page shows how long that live query took.

Demonstration only. The demo shows relationships recorded in public filings; nothing in it indicates or suggests wrongdoing by any person or organization. Individual donors appear only under serial pseudonyms. The live trace returns when the demo comes back online; until then this section describes it.

  1. Load the filings. FEC contributions and committee spending for the 2022, 2024 and 2026 cycles, IRS Form 990 grant schedules and officer lists, and the ICIJ Offshore Leaks graph.
  2. Record provenance. Every fact keeps the source file it came from and the snapshot it belongs to, so a result can be traced back to the filing.
  3. Link across sources. A committee, nonprofit or company that appears in two datasets is joined only on an exact match, and the link records which fields matched.
  4. Pick a start. Choose any committee, nonprofit or company and a depth of up to six hops.
  5. Watch it expand. For each hop the engine reports the frontier size, the entities reached so far and the connections explored, measured live for that query.