Frequently asked questions.
Do I need to rewrite my Spark code?
No. Quanton is a drop-in replacement for the Spark execution engine. Your existing code, APIs, and pipelines work exactly as before.
What versions of Apache Spark are supported?
Quanton supports Spark 3.x. See the docs for the full compatibility matrix.
How does Quanton achieve faster performance?
Quanton uses vectorized, columnar execution with SIMD acceleration — optimized and reimplemented Velox operators — plus storage-aware planning: native reads with I/O overlapped with compute, index-aware joins, low-shuffle MERGE, and native clustering/compaction for Hudi and Iceberg.
How much faster is Quanton than Apache Spark?
Up to 5× on TPC-DS 10 TB: open-source Spark takes 12,200 seconds across all 99 queries vs Quanton's 2,384 seconds on the same 11-node cluster. Stage-level gains include ~7× lower scan time (TPC-DS q23), ~6× faster MERGE INTO at 1 TB, and ~4× faster clustering and compaction.
Is Quanton faster than Databricks Photon?
Yes, slightly — on TPC-DS 10 TB total runtime, Quanton finishes in 2,384 seconds vs Photon's 2,550 seconds (6.5% faster) on identical hardware, with no per-cluster premium. On that run Quanton's per-GiB fee was $5.27 vs $15.67 in Photon DBU fees — 66% lower.
How does Quanton compare to Apache Gluten and DataFusion Comet?
On TPC-DS 10 TB, Quanton is 3.6–3.8× faster than tuned Gluten (8,563s) and Comet (9,122s). Both OOM on q67 and q93 even after tuning; Quanton clears all 99 queries and degrades gracefully to vanilla Spark instead of failing under memory pressure.
Where does Quanton run?
Quanton runs anywhere Spark runs — on-prem, AWS, GCP, Azure, or any Kubernetes environment.
How much does Quanton cost?
Quanton is priced per GiB of data processed, not per compute hour — the first 100 GiB each month is free, and tiered rates start at $0.0009/GiB read. Faster releases cut your EC2 hours without raising your Quanton bill, so the savings stay with you.
What is the Quanton AI agent for Spark?
An AI agent built into the Spark Web UI that diagnoses slow, failing, or expensive jobs — data skew, shuffle bottlenecks, OOM risk, GC pressure, executor underutilization, and partition sizing — and recommends config or code fixes. It works from the live job context: stages, tasks, SQL plans, executor logs, and alerts, with no job code changes.
Is Quanton open source?
Quanton has an open core. See the docs and GitHub for details on what's open and what's available under a commercial license.
How do I get started?
Head to the docs for a quickstart guide. You can be up and running in under 10 minutes.
