Not one estate in the review file held a maintained concurrency baseline, so the number that decided the license position was the authorized account list
Concurrency is not a property of the estate. It is a property of the evidence, and the party that reconstructs it first with a stated method controls everything after.
Prepared by Redress Compliance · August 19, 2026 · JD Edwards concurrency reviews. 20 to 30 estates, 2024 to 2025.
Executive summary
Zero estates held a baseline. Across roughly 20 to 30 JD Edwards estates reviewed in 2024 and 2025, a minority still held concurrent quantities and not one of them had a maintained concurrency baseline.
Default log retention runs 7 to 30 days against an audit lookback measured in years, so the surviving evidence covers a fortnight and usually includes a period end spike.
Machine sessions sit inside the human figure. Batch, interface and monitoring identities hold application connections the same way a person does, adding a double digit percentage to the apparent peak.
With no data, the fallback is the account list. That is not a contractual rule. It is what happens when one side has evidence and the other has an assertion.
What does a concurrent license actually count?
It counts the largest number of qualifying sessions live at the same instant inside a defined measurement period. Three variables sit inside that sentence, and the contract sets all three.
What qualifies as a session, how long the period runs, and what counts as the same instant. Get any one wrong and the number changes materially. Oracle documents the product on its JD Edwards EnterpriseOne page, but the metric wording lives only in the ordering document.
What counts as one session
- A signed in human at one workstation. Uncontroversial, and rarely the issue.
- The same human in three browser tabs, which the wording may treat as one connection or three.
- A process holding an application connection. A log cannot tell it apart from a person without help.
- A connection left open by a closed browser. Sessions end when the timeout says they do.
Which four phrases in the order decide the number
- The unit. Concurrent user, concurrent device and concurrent session are three different counts.
- The qualifier. Some wordings count only sessions performing a transaction, others any active connection.
- The scope. Per component, per instance or across the enterprise changes everything downstream.
- The measurement clause. If the order states a method, that method binds both sides. If it is silent, propose one first.
Where do surviving concurrent quantities come from?
Older paper, in every case. Oracle's published price lists moved to the application user and employee metrics long ago, so concurrent quantities in live estates arrived before that shift.
Check the current price lists and your own order rather than taking anybody's word for it.
Three routes it survived by
- Pre acquisition agreements signed with J.D. Edwards or PeopleSoft, carried forward through assignment.
- Early Oracle era orders that preserved the incumbent metric to close a deal quickly.
- World customers whose paper reflects a different product generation and counting tradition.
Legacy paper is usually an asset. It is also fragile, because it rests on documents that are decades old and on definitions Oracle has no commercial interest in reaffirming.
A new purchase is the moment the legacy metric comes under pressure, and there is no published conversion ratio. It is derived from evidence, so if the only evidence in the room is a list of named accounts, that list becomes the ratio.
- The unit, the qualifier, the scope and the measurement clause, quoted back with the page
- Where the wording leaves method negotiable, and what to propose first
- Paste ready language for the measurement clause your order is missing
How is a concurrent peak actually measured?
It is reconstructed, never read off a dial. JD Edwards has no single counter that produces a defensible concurrency figure.
Somebody assembles one from several systems, and the assembly method decides the answer as much as the estate does.
| Source | What it shows | Typical retention | Evidential weight |
|---|---|---|---|
| Web and application tier server records | Live user sessions on the presentation layer | Days to weeks by default | High, closest to a human |
| Load balancer or reverse proxy logs | Connection counts by time and source | Often 30 days | Medium, no identity |
| Sign on auditing, where switched on | Who signed in and when | Whatever you configured | High, but rarely enabled in advance |
| Batch job records | Server side jobs and their run windows | Purged on a schedule | High for excluding machine load |
| Database session views | Connections, including pooled ones | Instantaneous unless sampled | Low, pooling hides the human count |
Read the retention column again. An audit request reaches back further than default log settings survive, which is the whole problem in one line.
How the sampling method changes the answer
- Instantaneous peak. The highest single reading anywhere in the period, volatile, and the number an auditor prefers.
- Sustained peak. The highest reading held for a stated duration, far closer to what the business needs.
- Percentile. Statistically honest, and worth proposing where the contract is silent on method.
- Business day filter. Whether overnight batch sits inside the measured period is a decision, not a fact.
When does concurrent beat the application user metric?
When the ratio of authorized individuals to defensible peak is larger than the price multiple between the two metrics. That is the whole test.
It is arithmetic rather than judgement, once both numbers exist.
A worked build from a raw reading
| Step | Adjustment | Sessions |
|---|---|---|
| Raw instantaneous maximum, period end Tuesday | Starting reading | 260 |
| Remove integration and interface identities | Minus 34 | 226 |
| Remove monitoring and availability probes | Minus 22 | 204 |
| Remove connections idle beyond the stated policy | Minus 58 | 146 |
| Compare with authorized individuals on the estate | 900 people | Ratio of about 6 to 1 |
Composite figures drawn from the pattern across engagements, not a single client. Put your own price list against it.
If a concurrent unit costs less than six times an application user for the same component, concurrent wins at this ratio. If it costs more, the legacy metric is sentiment rather than strategy.
Which metric suits which access pattern
| Access pattern | Ratio to expect | Metric that usually wins |
|---|---|---|
| Three shifts, rotating, plant floor | 4 to 1 or wider | Concurrent, if you can prove it |
| Office hours finance and procurement | Around 2 to 1 | Application user |
| Seasonal peaks, thin baseline | Varies wildly by month | Concurrent, period argued first |
| Heavy machine and interface traffic | Distorted by non human sessions | Application user |
| Workforce modules touching everybody | Not applicable | Employee metric |
What does Oracle ask for when concurrency is in scope?
Something broader than concurrency, almost always. The request that lands is a standard applications data collection, and concurrency is not what it is designed to produce.
That puts you in the position of volunteering the only evidence that helps you. License Management Services is not obliged to accept an estimate you cannot reproduce.
Four of the five items build a named user count
- A full extract of user profiles and security records, which produces a headcount rather than a peak.
- Role and menu assignments mapped to components, which produces authorization rather than simultaneity.
- Environment and instance inventory, including training, disaster recovery and the migration test copy.
- Installed component evidence, which sets the scope of everything that follows.
- Any measurement output you hold, the only line where your baseline can enter the record.
Read that list from a negotiating position. Four of the five build the case for a named user count, and the fifth is optional unless you supply it.
Where the common advice on JD Edwards concurrent licensing is wrong
The common advice is to defend the concurrent metric at all costs because it is almost always cheaper than named licensing. That advice is right about the price and wrong about the risk.
A metric you cannot measure is a liability, not a saving. If logs rotate in a fortnight, sign on auditing was never switched on, and the paper defines no measurement method, the concurrent quantity is an unpriced option Oracle can call at a moment of its choosing.
Converting from strength costs less than defending from ignorance. The second route is the one most estates take by default, because nobody chose the first one in time.
None of the 20 to 30 estates reviewed in 2024 and 2025 held one.
Against an audit lookback measured in years, not weeks.
900 authorized individuals against a defensible peak of 146.
The Oracle audit defense checklist
The evidence sequence, the exclusions that survive scrutiny, and the wording that decides the count.
Read the metric guide →What 20 to 30 JD Edwards estates showed
A minority of the estates reviewed in 2024 and 2025 still held concurrent quantities. Not one of them had a maintained concurrency baseline.
The four patterns that repeated
- The concurrent quantity was inherited from a contract older than everyone in the room, and the definition of session had never been read aloud.
- Web tier logs had rotated away, so surviving evidence covered the previous fortnight and included a period end spike.
- Machine sessions, monitoring probes and integration accounts sat inside the same figure as human users.
- Where a conversion had already happened, nobody could explain how the ratio was arrived at, because the concurrency evidence had never existed.
Concurrency is not a property of your estate. It is a property of your evidence, and changing the evidence changes the license position entirely lawfully.
The metric definitions behind all of this are set out in our JD Edwards user types and metrics guide, and the same evidence discipline applies across the Oracle EBS licensing series.
Research briefingApplication users, and what the metric actually countsWhy the application user count and the authorized account list are not the same number, and which one an auditor reaches for first.
Your first five moves
- Find the four phrases in your ordering document before accepting anybody's arithmetic, including your own.
- Extend log retention now, because the cheapest evidence is the evidence you already had.
- Build a dated concurrency series with the method written down first: interval, period, inclusion rules, exclusions.
- Name every excluded machine identity. A removed session that cannot be named is an argument rather than a fact.
- Decide deliberately whether to defend the metric or trade it, and trade it only when Oracle wants to sell you something.
Frequently asked questions
What does a JD Edwards concurrent license count?
The largest number of qualifying sessions live at the same instant inside a defined period. The contract sets what qualifies, how long the period runs, and what counts as the same instant.
How many estates held a concurrency baseline?
None. Across roughly 20 to 30 estates reviewed in 2024 and 2025, a minority still held concurrent quantities and not one had a maintained baseline.
Why does log retention matter so much?
Because default retention runs 7 to 30 days while an audit lookback runs years. The surviving evidence covers a fortnight and usually includes a period end spike.
Do machine sessions count toward the peak?
They sit inside the figure unless you remove them. Batch, interface and monitoring identities hold connections like people do and add a double digit percentage.
What happens with no concurrency data at all?
The fallback becomes the number of authorized accounts, because that is the only figure both parties can see and it is the highest one available.
Can you still buy concurrent licensing?
Not from the current component price list, in our reading of it. Surviving quantities arrived on older paper, so check the published lists and your own order.
When does concurrent beat application user?
When the ratio of authorized individuals to defensible peak exceeds the price multiple between the metrics. In the worked build that ratio was about 6 to 1.
What is the conversion ratio at a new purchase?
There is no published one. It is derived from evidence, so a list of named accounts becomes the ratio if that is all the evidence in the room.
Which sampling method should you propose?
A sustained peak or a percentile where the contract is silent, because both sit closer to what the business needs than a single volatile reading.
Is defending the metric always right?
No. A metric you cannot measure is a liability rather than a saving, and converting from strength costs less than defending from ignorance.