Mage Data and Perforce Delphix both deliver protected test data, and they came at the problem from opposite directions. Delphix started with database virtualization and added masking. Mage Data started with data discovery and protection, and added delivery. That difference decides where each one fits.
How does Perforce Delphix approach test data management?
Perforce Delphix is a DevOps data platform built around database virtualization. It creates virtual copies of databases from shared data blocks, so teams get full-size working environments in minutes instead of waiting for physical copies. Perforce completed its acquisition of Delphix in March 2024, and the product now sits alongside Perforce’s wider DevOps portfolio.
Delphix groups four products under test data management. Continuous Data handles virtual delivery, with refresh, rewind, bookmark and branch operations that let teams move an environment to a known data state. Continuous Compliance handles discovery and masking, using templates built for specific applications and regulations, then applying irreversible masking algorithms or tokenization. Delphix Synthetic Data generates data when production is unavailable. Data Control Tower provides the catalogue, access control and API gateway across the other three.
How does Mage Data approach test data management?
Mage Data is a data security platform. Test data management is one capability within it, alongside production data protection and compliance automation. The platform starts with discovery: a patented, score-based engine classifies sensitive data across databases, files, data lakes, warehouses, logs and packaged applications, including content inside free-text fields.
A policy maps each classification to a protection method, and that policy then governs every way data is delivered. Test data management covers in-place masking of an existing copy, extract-mask-load pipelines that protect data in transit, file gateways that mask files as they land in object storage, and database virtualization serving writable copies from one masked golden snapshot. Masking is deterministic across systems, so the same customer looks the same everywhere and joins still work.
Is Delphix a test data management tool or a database virtualization tool?
Both, and the order it was built in still matters. Delphix began as database virtualization and added masking later, which is why the two ship as separate products within the platform rather than as one.
For an evaluation this has two practical consequences. First, check which product carries the capability you need and whether both are licensed. Second, when a capability sits in a separate product, it brings its own configuration and its own audit trail, so the number of places a protection rule has to be defined goes up.
Mage Data takes the opposite approach by design. Classification and protection are defined once, and virtualization is one of several delivery paths that enforce the same policy.
Key differences at a glance
Perforce Delphix
- Virtualizes databases and serves virtual copies from shared blocks, with refresh, rewind, bookmark and branch.
- Masking ships as a separate product within the platform, Continuous Compliance.
- Discovery works from templates built for specific applications and regulations.
- Subsetting is not central to the approach, because virtualization makes full copies cheap.
- Strongest where the estate is relational and environment lead time is the constraint.
Mage Data
- Discovery runs first, across databases, files, lakes, warehouses, logs and packaged applications, including free text.
- One policy maps each classification to a protection method, and every delivery path enforces it.
- Four delivery paths plus virtualization: in-place, extract-mask-load, file gateways, self-service.
- Intelligent Subsetting from a single driver condition, with referential integrity resolved automatically.
- Strongest where sensitive data is spread beyond databases and the control has to be provable.
How do Delphix and Mage Data differ?
| Mage Data | Perforce Delphix | |
|---|---|---|
| Origin | Data security platform | Database virtualization |
| Database virtualization | Yes, masked at snapshot creation | Yes, with refresh, rewind, bookmark and branch |
| Sensitive data discovery | Score-based across structured, free-text and unstructured content | Discovery templates for specific applications and regulations |
| Static masking | Irreversible, method selected per classification | Irreversible algorithms and tokenization |
| Data subsetting | Intelligent Subsetting from one driver condition | Not central to the approach |
| Delivery paths under one policy | In-place, extract-mask-load, file gateways, virtualization | Virtualization and masking as separate products |
| Synthetic data | Mage Data’s Identities builds a consistent universe of fictitious people. The same customer becomes the same fictitious customer in every system, with demographics and distributions preserved, so cross-system tests and analytics still behave. Further synthetic data capabilities are on Mage Data’s roadmap, including generation from a schema alone and full digital-twin datasets. | Synthetic values via Delphix AI |
| Scope | Test data as one capability of a data security platform | Test data and DevOps data management |
| Ownership | MENTIS Inc, independent | Perforce Software, since March 2024 |
Both platforms mask irreversibly. The difference is who the masked copy is for.
One masked copy serves one kind of requester. Mage Data masks the same table differently depending on who asked and why. A developer gets realistic data with the context preserved, so the application behaves normally. A DBA gets the database with sensitive values redacted. An acceptance tester gets some real values and the rest de-identified, referentially intact. A model pipeline gets the demographics and distributions kept and the identities removed, in the format it needs.
Same policy, same classification, different output per consumer. That is fit-for-purpose masking: the data is useful for the job it was requested for, and useless for anything else.
See it on your own data
Discovery is the part of this comparison you can test in an afternoon. Send us one source system and we will show you what classification finds in it, including the false positives. Book a demo →
Does Delphix do data subsetting?
Not as a central part of its approach, and that is a design choice rather than a gap. The logic is that virtualization makes full-size copies cheap enough that shrinking them becomes less necessary. Bloor Research has made the same observation in its test data management research.
The logic holds until a team needs a small targeted dataset for a specific test, or a third party can only receive the minimum necessary records, or a data minimisation obligation applies to what leaves production at all.
Mage Data’s Intelligent Subsetting takes one driver condition, such as customers in a single region, and resolves upstream and downstream referential integrity automatically. Subsetting can also run in transit, as part of an extract-mask-load pipeline, so the smaller dataset is produced on the way to its destination rather than afterwards.
How does Delphix handle files and unstructured data?
Virtualization operates on databases, so for data that sits outside a database the question is what governs it and whether the same policy applies. This matters more than it used to, because a growing share of sensitive content in non-production environments is in files, extracts and documents rather than in tables.
Delphix’s masking product, Continuous Compliance, identifies sensitive data using discovery templates and masks it. For a file estate, ask specifically which formats are covered, how files arriving continuously are handled, and whether that path shares a policy with the database masking.
Mage Data runs file gateways that detect files landing in object storage or a directory, mask their contents and forward them, under the same policy that governs database masking. Discovery also reads inside free-text fields, which is where account numbers and identifiers often end up after a support call.
What changed after Perforce acquired Delphix?
Perforce completed the acquisition on 25 March 2024, and has continued to develop the platform alongside its wider DevOps portfolio. The products and their names have carried over.
For a customer the useful check is the same as for any ownership change, and it does not depend on anyone’s prediction. Compare release notes for the twelve months before and after. Check whether support lifecycle announcements have changed in tone or cadence. Look at what moved in your renewal terms. If none of those have shifted against you, the acquisition is not a reason to move.
One thing worth confirming directly: whether virtualization and masking are still licensed as separate products or have been bundled. Licensing structures often change quietly after an acquisition.
When should you choose Delphix, and when should you choose Mage Data?
Both platforms deliver protected test data. The decision usually comes down to where your sensitive data actually sits and what is slowing your teams down.
| Choose Perforce Delphix when | Choose Mage Data when |
|---|---|
| Your estate is mostly relational databases and the pressing problem is how long a full-size environment takes to arrive. | Sensitive data also sits in files, extracts, documents and free-text fields, and you need one policy to cover all of it. |
| Teams need to bookmark a data state, reproduce a defect against it, then rewind. Data versioning is part of how they work. | An auditor asks what was in a specific environment on a specific date, and how it was protected. You need a record you can export and hand over, not a dashboard someone has to be walked through. |
| You have standardised on Perforce for the rest of the DevOps toolchain. | You are consolidating discovery, masking and provisioning rather than buying another point tool. |
| Full copies are what testing needs, and subsetting is not a requirement. | Mage Data delivers protected data through Secure Data Provisioning, which masks data as it moves from source to target; in-place Static Data Masking for a copy that already exists; Intelligent Subsetting for a smaller, referentially intact slice; Secure File Gateways for files landing in object storage; and Database Virtualization for writable copies from one masked snapshot. All of them run from the same policy, so the route is chosen per request rather than fixed by the platform. |
| Virtualization performance across a defined list of database platforms is the main technical constraint. | Discovery has to find sensitive data nobody labelled, including inside free text and key-value structures. |
If you already run Delphix, the question is usually not replacement. It is what sits outside the virtualized databases, and who governs that today. Where that list is short, Delphix covers the problem. Where it is long, the gap is worth measuring before the next renewal.
What should you test before choosing?
Both vendors will demo well. Every platform works on data it has been tuned against, which is why the only useful test is one run on your own schema, with your own awkward cases in it. Six tests, and what each one tells you.
Put an account number inside a free-text field, then run discovery.
You are testing whether classification reads content or only column names and patterns. Ask for the false positives too. A platform that finds everything by flagging everything has not solved the problem.
Mask one customer identifier and trace it.
Show the same value in a database table, a file in object storage, and an extract landing in your warehouse. If it matches in all three, cross-system testing still works after protection. If it only matches inside the database, every join that crosses a boundary becomes your problem.
Generalize a numeric field, then check the arithmetic.
Round a quantity into a band so it cannot be matched against public data, and see whether the fields derived from it move with it. This separates the field fastest. If the answer involves a services engagement or a custom script, you have your answer.
Count the places one rule has to be defined.
Take a single protection rule and ask how many times it must be configured to cover every delivery path you actually use. One is a platform. Four is four tools with a shared logo.
Time a provisioning run, including the approval.
Use a database of your size, not a reference dataset, and start the clock where a requester would start it. The number that matters is request to usable environment, not job runtime.
Ask for the audit record, not the dashboard.
Name an environment and a date, and ask what the platform can produce about what was in it and how it was protected. An assessor needs an exportable artefact. A screen is not one.
The TDM evaluation guide
These six tests are the short version. The full guide gives your buying committee six weighted scoring dimensions, twelve demo questions with strong and weak answers, and a proof-of-concept checklist for the stage where the decision actually gets made. Get the TDM evaluation guide →
The bottom line
Delphix solved a real problem and solved it first. If your sensitive data lives in relational databases and your teams are waiting on environments, that is still the problem it solves best.
Two questions decide whether it is still the problem you have.
Is the copy fit for the purpose it was requested for? A fast copy is one copy, shaped one way. A developer, a DBA, an acceptance tester and a model pipeline need different things from the same table — realistic data with context preserved, values redacted, some real values and the rest de-identified, distributions kept and identities removed. Fit for purpose means each of them gets the shape their work needs, from the same policy. When the copy doesn’t fit, teams go back to production. That is the quiet failure, and it rarely appears in a risk register because it looks like a delivery problem.
And can you get it there the way your teams actually work? Not every request suits the same route. Some need a copy masked where it already sits, some need it masked as it moves, some need a slice rather than the whole database, some need files protected as they land. Virtualization answers one of those well. The rest is where most non-production exposure accumulates.
So the comparison comes down to this: how much of your non-production estate sits outside what virtualization covers, whether what arrives is fit for the purpose it was asked for, and whether one policy reaches all of it.
You can work that out without us.
- List your non-production environments.
- Mark which hold data from production.
- Mark which are covered by a policy you could show an auditor.
- If the second number is much larger than the third, that gap is your actual problem, and it is worth sizing before your next renewal.
Book a demo
Thirty minutes. We’ll show you the same table delivered four ways — to a developer, a DBA, an acceptance tester and a model pipeline — from one policy, and the record of what was done to it.
Frequently asked questions
What is test data management?
Test data management is the practice of finding sensitive data, protecting it, and delivering realistic, referentially intact copies to the teams and systems that build and test software. It now covers more than developers and QA. The same production-shaped data feeds performance testing, analytics, AI training pipelines, third-party processors and auditors, and each needs a different shape of it.
What is data virtualization in test data management?
Data virtualization creates lightweight, writable copies of a database that share storage with one master copy, so each team gets what behaves like a full database without storing a full copy. It reduces storage cost and provisioning time. It does not, on its own, decide what in the copy is sensitive or whether it has been protected, which is why virtualization and masking are usually separate capabilities.
What are the main alternatives to Perforce Delphix?
The alternatives most often evaluated for test data management are Mage Data, K2view, Informatica, IBM InfoSphere Optim, Broadcom Test Data Manager and Tonic.ai. They differ mainly in whether they provision whole databases or targeted subsets, how broadly they discover sensitive data, and how many delivery paths run from a single policy.
When should you look at a Delphix alternative?
When sensitive data sits outside relational databases, in files, extracts, documents or free-text fields. When testers need targeted subsets rather than full copies. When a third party can only receive the minimum necessary records. Or when you need one exportable record showing what was in a named environment on a named date, covering every delivery path rather than the virtualized databases alone.
Can Delphix and Mage Data be used together?
Yes, and some organizations do run virtualization for relational environments alongside a separate control for files and extracts. It works. It also means two policy definitions and two audit trails to reconcile, so the cost worth quantifying is the reconciliation effort, not the licences. Where that cost is low, coexistence is reasonable. Where it is high, it is usually a stage rather than a destination.
How does test data management support PCI DSS and GDPR compliance?
PCI DSS v4.0.1 Requirement 6.5.5 bars live primary account numbers from pre-production environments, unless that environment is included in the cardholder data environment and protected under all applicable PCI DSS requirements. Under GDPR, the EDPB’s Guidelines 01/2025 on Pseudonymisation confirm that pseudonymized data which can still be attributed to a person using additional information remains personal data. In practice, both point to the same requirement: protection applied before the copy exists, and a record that shows it.
What about data used to train AI models?
Training and evaluation sets are copies of production like any other, and they tend to be the least governed because they are created by engineering activity rather than by a business process. The control point is extraction, before the data reaches the pipeline, because a model cannot be made to unlearn what it was trained on without retraining. Treat an AI training set as another consumer of test data, governed by the same policy.
Does masking slow down provisioning?
It depends which route the data takes, and Mage Data offers three.
In-place masking protects a copy that already exists, where it sits. The copy is created first, so there is an interval before masking runs during which real data sits in non-production. Simplest to adopt, and the only one of the three with that window.
Secure Data Provisioning masks data in transit, as it moves from source to target. Nothing unprotected lands at the destination, and Intelligent Subsetting can run on the way so the target gets only the records it needs.
Database Virtualization masks once, when the shared copy is created. Every virtual copy served from it is protected before it exists, so masking is not in the path of any individual request. This is where provisioning drops to minutes.
The question worth putting to any vendor is which of these they support, and whether all of them run from the same policy or each needs configuring separately.
Last updated November 2026 · This comparison is based on public product documentation as of November 2026. Perforce and Delphix are trademarks of Perforce Software, Inc. Mage Data is not affiliated with Perforce. Mage Data offers a competing product in the test data management category.