GraphForge vs. JanusGraph
What do pluggable distributed storage backends buy you?
JanusGraph is a Linux Foundation property graph database built on Apache TinkerPop. Graphs are stored in Cassandra, HBase, ScyllaDB, Bigtable, or BerkeleyDB, and indexed by Elasticsearch, Solr, or Lucene. Analytics run as TinkerPop VertexPrograms on Spark. The design scales to clusters you already operate, at the cost of operating them. GraphForge needs no backend at all, runs Cypher in-process, and keeps 94 algorithms in the engine.
On this page
Should I choose JanusGraph or GraphForge?
JanusGraph to store a huge graph on Cassandra or HBase; GraphForge to analyze one without running a cluster.
Choose GraphForge when
- You want no storage or index cluster to run for analysis.
- You need algorithms in the engine rather than Spark OLAP jobs.
- You want durable atomic local writes rather than transaction semantics that vary by storage backend.
Choose JanusGraph when
- You already operate Cassandra, HBase, ScyllaDB, or Bigtable and want the graph stored there.
- The graph must scale across a storage cluster, up to 2^60 edges by its ID scheme.
- Your team uses Gremlin and the TinkerPop ecosystem.
What can each one do?
20 of 26 capabilities differ. For 12, JanusGraph's docs did not confirm support, so they are marked Not established rather than No.
Each capability has a status and a source. “Not established” means we could not confirm the feature from JanusGraph's primary documentation. It does not mean JanusGraph lacks it.
| Capability | GraphForge | JanusGraph |
|---|---|---|
| Storage | ||
| Property graph model | Shipped Labeled property graph with typed properties. 1 | Yes Native property graph model via Apache TinkerPop. 1920 |
| Document model | Not offered No document-store model; properties accept nested, heterogeneous values. 1 | Not established No document data model documented. |
| Relational tables | Not offered Not a relational database; results arrive as Arrow tables. 1 | Not established No SQL or relational interface documented. |
| Key-value access | Not offered No key-value API; data is a property graph. 1 | Not established Persists to wide-column/KV backends, but exposes no key-value API. |
| Columnar storage | Shipped Durable projects store graph data as Arrow and Parquet. 4 | Not established No columnar analytics storage documented. |
| Vector search | Shipped Exact cosine retrieval over embeddings you supply; no ANN index. 5 | Not established No vector index or similarity predicate in documented search predicates. |
| Full-text search | Shipped Tantivy-backed BM25; hybrid text and vector retrieval through find(). 5 | Extension Text predicates (textContains, fuzzy, phrase) need a mixed index on Elasticsearch, Solr, or Lucene. 2320 |
| Geospatial | Not offered No spatial types or indexes are documented. | Extension Geo predicates (geoWithin, geoIntersect, etc.) served by external mixed-index backends. 23 |
| Execution | ||
| Transactional workloads (OLTP) | Partial Local analytical scope; durable atomic writes with queued and optimistic writer modes. 6 | Yes Transactional database for thousands of concurrent users and complex traversals. 1921 |
| Analytical workloads (OLAP) | Shipped Analytical verbs and Cypher over a local columnar project. 7 | Yes Global analytics through TinkerPop hadoop-gremlin/Spark, reading Cassandra or HBase. 22 |
| Graph traversal | Shipped Cypher MATCH patterns plus BFS, DFS, Dijkstra, A*, and reachability. 8 | Yes Gremlin traversals executed embedded or via JanusGraph Server (Gremlin Server). 20 |
| Variable-length paths | Shipped Bounded and unbounded patterns and shortestPath(). 8 | Yes Multi-hop traversal via TinkerPop Gremlin repeat()/until() steps. 26 |
| Graph algorithms | Shipped 94 native algorithms behind analysis verbs, not Cypher CALL procedures. 37 | Partial Only TinkerPop's generic VertexPrograms (PageRank, PeerPressure, components, shortest path); no dedicated library. 26 |
| Incremental computation | Not offered Algorithms recompute on request; no incremental mode is documented. 7 | Not established No incremental algorithm maintenance documented. |
| Deployment | ||
| Embedded / in-process | Shipped In-process library; no server, daemon, or connection string. 9 | Yes Can be embedded in a JVM application; queries, caches, transactions run in-process. 20 |
| Server mode | Planned Optional remote peer extension planned after 0.6.0. 11 | Yes JanusGraph Server builds on TinkerPop Gremlin Server; stateless, load-balanceable instances. 2025 |
| WebAssembly | Not planned A browser WebAssembly engine was declined in ADR 0027. 12 | Not established No WebAssembly build documented. |
| Distributed cluster | Not planned Distributed clusters and HA replication are out of scope. 9 | Yes Graph partitioned across a multi-machine storage cluster (Cassandra, HBase, ScyllaDB, Bigtable). 1920 |
| Managed cloud | Not planned No hosted compute. The graphforge.sh Hub distributes repositories; execution stays local. 13 | Not established No official managed JanusGraph service; Bigtable is managed storage only. |
| Investigation | ||
| Branching | Experimental Research versions, branches, slices, proposals, and forks ship in 0.6.0 release candidates. 14 | Not established No graph branching documented. |
| Snapshots / time travel | Shipped Named checkpoints you can diff and revert, plus transaction-time snapshots. 3 | Not established No point-in-time historical queries documented. |
| Immutable history | Shipped Knowledge ledger records immutable assertions and evidence lineage. 16 | Not established No immutable history; optional transaction log records changes only when enabled per transaction. |
| Provenance | Shipped Provenance and knowledge ledger with evidence lineage. 16 | Partial Opt-in user transaction log records changes per named log; not full lineage. 27 |
| Competing hypotheses | Shipped Immutable hypothesis groups with membership and selection events. 3 | Not established |
| Operations | ||
| ACID transactions | Partial Acknowledged-durable atomic publication; docs do not claim general ACID or serializable isolation. 6 | Partial Transactions 'not necessarily ACID'; configurable on BerkeleyDB, not on Cassandra/HBase. 21 |
| Replication | Not planned HA replication is out of scope. 9 | Partial Delegated to the storage backend; homepage cites multi-datacenter HA and hot backups. 1925 |
Showing 26 of 26 capabilities.
How is each one built?
JanusGraph: property graph on pluggable backends. GraphForge: an embedded property graph library in Rust, persisted as Parquet.
GraphForge
GraphForge is a Rust library that runs inside your process: no daemon, server, or connection string. Projects persist to a local directory as Arrow and Parquet data plus JSON metadata, and every query and analysis verb returns an Apache Arrow table. Algorithms run in the same native engine as Cypher and are called as API verbs (rank, cluster, paths, analyze, similar), not as Cypher procedures.
JanusGraph
JanusGraph is a Java layer that maps a property graph onto a storage backend and delegates mixed indexes to a search engine. Servers are stateless and masterless behind a load balancer. OLAP runs through TinkerPop's hadoop-gremlin and Spark. Transaction guarantees depend on the backend.
JanusGraph is a set of distributed systems working together; GraphForge is one library. That matters for operational cost as much as for performance.
What the documentation says about JanusGraph
- JanusGraph has been community-driven under the Linux Foundation since 2017 and is Apache 2.0 licensed.19
- Supported storage backends include Apache Cassandra, Apache HBase, Google Cloud Bigtable, Oracle Berkeley DB Java Edition, and ScyllaDB.19
- Mixed indexes for full-text, geo, and numeric-range search are delegated to Elasticsearch, Apache Solr, or Apache Lucene.19
- Applications can embed JanusGraph in their JVM or send Gremlin to JanusGraph Server, which is built on TinkerPop Gremlin Server.20
- JanusGraph Server instances do not communicate with each other and have no master, so they scale behind a load balancer.25
- OLAP uses TinkerPop's hadoop-gremlin with Apache Spark, with input formats for Cassandra (CqlInputFormat) and HBase.22
- The ID scheme caps a graph at 2^60 edges and half as many vertices.24
- 1.1.0 (November 2024) was tested with TinkerPop 3.7.3, Cassandra 3.11/4.0, HBase 2.6.0, ScyllaDB 6.2.0, and Elasticsearch up to 8.15.3.17
- The unreleased 1.2.0 changelog drops Java 8, moves to TinkerPop 3.8.2, and adds Cassandra 5, Elasticsearch 9, and OpenSearch support.17
Which one is faster?
Not measured yet. No head-to-head run has been published, so this page shows no performance chart.
Charts appear only after a reproducible run exists: pinned versions, disclosed hardware, generated data, raw results, and scripts.
No comparative results yet
The suites below apply to GraphForge and JanusGraph. Each will be reported separately per benchmark class, never merged into one leaderboard.
Ingestion and writes
| Suite | Workloads | Metrics | Status |
|---|---|---|---|
| INGEST | CSV, JSONL, Parquet, graph edge lists | Records/s, MB/s, total elapsed | Not run |
| MUTATE | Insert, update, delete, upsert | Ops/s, p50/p95/p99 latency | Not run |
Query, analytics, and system
| Suite | Workloads | Metrics | Status |
|---|---|---|---|
| LOOKUP | ID lookup, indexed property filter | p50/p95/p99 latency | Not run |
| TRAVERSE | BFS, k-hop, variable-length path | Latency, visited edges/s | Not run |
| ALGO | PageRank, WCC, SCC, Louvain, SSSP | Time, memory, correctness | Not run |
| SCALE | Increasing graph size and density | Throughput degradation, limits | Not run |
| CONCUR | Mixed readers and writers | Throughput, tail latency | Not run |
Which graph algorithms ship?
GraphForge: 94 native algorithms. JanusGraph: 4 named in its docs, delivered as bundled.
Algorithms are listed by name, as documented. Counts of extension modules or third-party libraries are not presented as native coverage.
GraphForge
94 Rust algorithms run in the engine behind five analysis verbs (rank, cluster, paths, analyze, similar) and return Arrow tables.37
A selection of the catalog:
- PageRank
- ArticleRank
- Betweenness (Brandes)
- Closeness
- Harmonic closeness
- Eigenvector
- HITS
- CELF influence maximization
- Triangle count
- K-core decomposition
- Louvain
- Leiden
- Label propagation
- Infomap
- Walktrap
- Girvan-Newman
- Weakly connected components
- Strongly connected components
- Dijkstra
- A*
- Bellman-Ford
- Yen's k-shortest paths
- Max flow / min cut
- Steiner tree
- Minimum spanning tree
- Topological sort
- Node2Vec
- GraphSAGE
- Fast random projection
- Node similarity (Jaccard)
JanusGraph
No native algorithm library; analytics run as Apache TinkerPop VertexPrograms/GraphComputer steps (typically SparkGraphComputer) inherited from TinkerPop.2622
Delivered as: bundled
- PageRank (PageRankVertexProgram / pageRank())
- PeerPressure clustering (PeerPressureVertexProgram / peerPressure())
- Connected Components (ConnectedComponentVertexProgram / connectedComponent())
- Shortest Path (ShortestPathVertexProgram / shortestPath())
Where can each one run?
GraphForge runs in-process only. JanusGraph: embedded yes, server yes, cluster yes.
| Mode | GraphForge | JanusGraph |
|---|---|---|
| Embedded / in-process | Shipped In-process library; no server, daemon, or connection string. 9 | Yes Can be embedded in a JVM application; queries, caches, transactions run in-process. 20 |
| Server mode | Planned Optional remote peer extension planned after 0.6.0. 11 | Yes JanusGraph Server builds on TinkerPop Gremlin Server; stateless, load-balanceable instances. 2025 |
| WebAssembly | Not planned A browser WebAssembly engine was declined in ADR 0027. 12 | Not established No WebAssembly build documented. |
| Distributed cluster | Not planned Distributed clusters and HA replication are out of scope. 9 | Yes Graph partitioned across a multi-machine storage cluster (Cassandra, HBase, ScyllaDB, Bigtable). 1920 |
| Managed cloud | Not planned No hosted compute. The graphforge.sh Hub distributes repositories; execution stays local. 13 | Not established No official managed JanusGraph service; Bigtable is managed storage only. |
| Replication | Not planned HA replication is out of scope. 9 | Partial Delegated to the storage backend; homepage cites multi-datacenter HA and hot backups. 1925 |
What will my team write?
GraphForge: openCypher from Python, Node.js, Rust. JanusGraph: Gremlin (Apache TinkerPop).
JanusGraph developers write Gremlin through TinkerPop drivers or embed it in a JVM application, and must define a schema that cannot change once committed. GraphForge developers write openCypher from Python, Node.js, or Rust.
- Query languages
- GraphForge openCypher
- JanusGraph Gremlin (Apache TinkerPop)
- SDKs and bindings
- GraphForge Python, Node.js, Rust
- JanusGraph Java (embedded or driver), Python, .NET, Any Apache TinkerPop Gremlin driver
- Written in
- GraphForge Rust
- JanusGraph Java
- Surfaces
- GraphForge Python, Node.js, Rust crates, CLI, VS Code extension, Notebooks, Agent skills
What are the trade-offs?
JanusGraph's transactions vary by backend and analytics need Spark. GraphForge has no Gremlin or cluster.
A fair assessment cuts both ways.
GraphForge
- Single machine only; no distributed storage backend.
- No Gremlin support.
- No Java binding.
JanusGraph
- Transactions are not ACID on Cassandra or HBase, which lack serializable isolation and multi-row atomic writes.21
- Eventually consistent backends can yield 'phantom vertices' and rely on locking for uniqueness constraints.21
- Edge, property-key, and vertex-label definitions cannot be changed once committed; only renaming or new types are allowed.24
- Mixed indexes support only a subset of data types and no SET or LIST cardinality property keys.24
- Spark OLAP on YARN or Mesos needs extra undocumented configuration, and standalone clusters need JanusGraph jars copied to every executor.22
How would I migrate?
Load JanusGraph data through Arrow, pandas, or Polars. Gremlin queries are rewritten as Cypher.
Into GraphForge
- GraphForge 0.5.x loads data through Arrow, pandas, and Polars bulk calls or Cypher scripts; it has no native CSV loader.15
- Algorithms are API verbs (rank, cluster, paths, analyze, similar), not Cypher CALL procedures, so procedure calls become verb calls.7
- GraphForge passes the full openCypher 2024.3 TCK, so standard openCypher queries are the portable part of a migration.10
Moving from JanusGraph
- Gremlin is the only native query language; Cypher or SPARQL need third-party TinkerPop compilers that emit Gremlin.28
- Any TinkerPop-compatible Gremlin driver can connect, so Gremlin code from other TinkerPop systems is largely portable.28
- Releases pin tested backend and TinkerPop versions, so migrations must align Cassandra/HBase/Elasticsearch versions with the compatibility matrix.17
What else do people ask about JanusGraph?
Is JanusGraph still maintained? JanusGraph is maintained, transactions are not ACID on Cassandra or HBase, and Gremlin needs rewriting as Cypher.
Is JanusGraph still maintained?
Yes. Master branch commits continue in 2026. The latest release is 1.1.0 from November 2024, and 1.2.0 is listed in the changelog but not yet released.
Are JanusGraph transactions ACID?
It depends on the backend. JanusGraph's docs say transactions are not ACID on Cassandra or HBase, which lack serializable isolation and multi-row atomic writes.
Can I run Gremlin queries on GraphForge?
No. GraphForge speaks openCypher only, so Gremlin traversals need to be rewritten as Cypher patterns or path verbs.
How was this researched?
From 28 primary sources, accessed 2026-10-19, each cited where it is used.
Capabilities were checked against primary documentation, repositories, and release notes, accessed 2026-10-19. Each cell records two things: the capability’s state and the evidence behind it. Every GraphForge cell carries a release status. Shipped means stable 0.5.2; experimental means 0.6.0 release candidates only. GraphForge is pre-1.0 and makes no backward-compatibility guarantee yet.
- Yes documented natively
- Partial limited or caveated
- Extension official extension or module
- Paid edition paid or enterprise edition only
- Removed removed, with the version cited
- Not established not confirmed by our research; not a “no”
- Shipped in stable 0.5.2
- Partial shipped with documented limits
- Experimental in 0.6.0 release candidates only
- Planned on the roadmap
- Not offered outside the product's model
- Not planned explicitly out of scope
Benchmark classes
- Equivalent workloads. The same logical operation with the same correctness requirement on every engine.
- Native optimized workloads. Each engine's best supported implementation, including projections, indexes, extensions, and native algorithms.
- End-to-end application workloads. Time from raw data to a usable result: import, indexing, projection, computation, and output retrieval.
Sources · accessed 2026-10-19
- GraphForge repository README — GitHub
- GraphForge releases — GitHub
- GraphForge 0.5.0 changelog — GitHub
- Storage architecture — GraphForge Book
- Analytics integration: search and retrieval — GraphForge Guide
- ADR 0018: acknowledged durability and isolation — GitHub
- Algorithm verbs — GraphForge Book
- Cypher guide — GraphForge Guide
- openCypher compatibility — GraphForge Reference
- openCypher TCK compliance — GraphForge Reference
- Remote authority peer extension (#215) — GitHub
- ADR 0027: native runtime boundary — GitHub
- Release scope and roadmap — GraphForge Docs
- Keep research history — GraphForge Guide
- Graph construction — GraphForge Guide
- Knowledge public API v1 — GraphForge Book
- Changelog — JanusGraph Docs
- Commit history (master) — JanusGraph/janusgraph (GitHub)
- JanusGraph homepage — janusgraph.org
- Architectural Overview — JanusGraph Docs
- Transactions — JanusGraph Docs
- JanusGraph with TinkerPop's Hadoop-Gremlin — JanusGraph Docs
- Search Predicates and Mixed Index Data Types — JanusGraph Docs
- Technical Limitations — JanusGraph Docs
- Deployment Scenarios — JanusGraph Docs
- Apache TinkerPop Reference Documentation (GraphComputer, VertexPrograms, repeat step)
- Transaction Log — JanusGraph Docs
- Connecting to JanusGraph — JanusGraph Docs
Try GraphForge on your own graph.
Install the package, open a project, and run the same notebook we measure with.