Esc

↑↓ move↵ openIndex · Pagefind
Scheduling / Large schedules and time precision

Large schedules and time precision

Import, open, edit and calculate real P6 programmes of up to 30,000 activities, understand the size limits, lean loading, patch saves and measured timings, choose Auto, Days or Minutes time precision, and run long analyses as background jobs.

On this page

Programme schedules from P6 often run to ten or twenty thousand activities. The Planner, the engine and the import path are sized for them. This page lists the limits, what they cost in time, and how time precision decides whether a schedule is calculated in whole days or exact working minutes.

Limits

Limit Value
Uploaded source file (XER, P6 XML or Microsoft Project XML) 32 MiB
Activities per schedule 30,000
Relationships, or resource assignments, per schedule 150,000 each
Stored plan, including retained P6 metadata 48 MiB
Retained P6 metadata 8 MiB budget. It’s trimmed, then dropped with an import issue, when it doesn’t fit.
Import issues kept per import 5,001, beside the plan rather than inside it

At 600 to 1,000 bytes per activity, the 30,000-activity bound is reached before any byte limit, so it’s the one you’ll meet first. One exception is a heavily resource-loaded programme. The DCMA health check sends every assignment, so a 25,000-activity schedule with about two assignments per activity would be refused, while one with one assignment per activity fits.

Measured timings

Measured on a developer laptop through the same worker path the product uses. Treat them as orders of magnitude, not a service level.

Schedule Activities / relationships / assignments Import Calculate Health check
Synthetic export 10,000 / 20,476 / 0 2.1 s 1.4 s 1.1 s
Synthetic export 25,000 / 51,093 / 0 6.0 s 3.5 s 3.1 s
Real P6 programme 11,426 / 32,407 / 3,393 3.9 s 2.7 s 3.0 s
Real P6 programme (27 MiB XER) 19,202 / 52,826 / 35,467 10.2 s 5.0 s 7.4 s

Eight real projects that the previous release refused as too large now import and calculate. One more is still refused because its logic contains a loop. A loop is a real error in the schedule, not a size limit.

Import memory and other hot spots

Three costs were reworked in this release, each checked to give output identical to the code it replaced:

Measure Before Now
Peak import memory, real 27.3 MiB XER (19,202 activities) 1,128 MiB 563 MiB
Peak import memory, synthetic 25,000-activity export 862 MiB 317 MiB
Import time, the same 27.3 MiB XER (import step alone) 9.1 s 7.5 s
Group & Sort by WBS with totals, 25,000 activities 555–630 ms 48–90 ms
Change summary of one save, 30,000 activities with 300 edited 685–754 ms 126–162 ms, computed at save time

The import keeps only one parse of the file alive at a time and parses only the tables it keeps. Group & Sort reads each row’s fields once, and change summaries are stored with the version. See the change feed.

Working with a big plan

  • The grid and Gantt are windowed. Only the rows on screen are drawn, so scrolling a 20,000-row schedule stays smooth. Filters, Group & Sort and layouts work at full size.
  • Plans over 2,000 activities load lean and save as patches. See Very large plans in the Planner.
  • The largest files can lose their retained P6 metadata. If keeping it would push the stored plan past 48 MiB, it’s dropped and the import says so. That plan’s export then carries only the fields os.construction models. See P6 compatibility.
  • Level-of-effort and WBS-summary rows are dated by derivation. After the calculation, a level-of-effort activity spans the work it’s tied to, and a WBS summary spans its branch. These rows have no float and are never critical, and the calculation lists them so you can check them.
  • Import a big file once and work from the plan. Importing a 27 MiB XER takes about 10 seconds. Later work reads the adopted plan, not the file.

Very large plans in the Planner

Plans of up to 2,000 activities open and save exactly as before. Above that, the Planner changes how the plan travels between your browser and the server. Nothing changes in what you see or in the saved result.

  • Lean loading. The page sends only what the Planner’s views use. Fields no view reads, such as the early and late dates and floats stored in the imported P6 file, stay on the server, and rows are packed column by column. A notebook longer than 1,024 characters loads when you open its activity in Activity Details. Before editing codes, UDFs or notebook topics across the whole plan, choose Load activity notebooks.
  • Patch saves. Save and calculate sends only what you changed since the version you opened. The server applies it to the stored plan and saves through the same path as any save: the version check (a stale version is refused), the calculation and the history. Fields the browser never received, including the retained P6 metadata, are kept as stored. A patch save stores the same plan, and gets the same engine results, as sending the whole plan would.
  • Filters, sorts and group totals still run on the whole plan in the browser, and undo and redo work as usual.

Measured on synthetic schedules exported to XER and imported, through the product’s own code paths on a developer machine. Treat the times as orders of magnitude.

Activities 1,000 10,000 25,000
Opening the Planner, before → after 1.8 → 0.41 MiB 18.2 → 4.1 MiB 45.4 → 10.3 MiB
Saving a one-field edit, request before → after 1.3 MiB → 217 B 13.1 MiB → 219 B 32.8 MiB → 219 B
Save response, before → after 1.8 → 0.41 MiB 18.2 → 4.1 MiB 45.4 → 10.3 MiB
Building / unpacking the payload 16 / 6 ms 88 / 32 ms 214 / 88 ms
Grid projection / filter / WBS grouping with totals 7 / 0 / 22 ms 38 / 1 / 262 ms 82 / 1 / 687 ms

That’s about 430 bytes per activity in all, so a 30,000-activity plan opens with about 12 MiB, well inside the limits, without paging rows from the server. An automated test checks the patch rules, that a patch save stores the same plan as a full save, and a budget at 25,000 activities.

Time precision: Auto, Days or Minutes

P6 schedules in hours. A plan can be calculated in whole working days or in exact working minutes. You choose with Schedule › Settings › Time precision:

  • Auto (the default for every import and every new plan). The engine picks minutes when the schedule has sub-day information, and days otherwise. The Settings page tells you what it found, for example Minutes (auto-detected: 94 activities with timed planned dates, 2 sub-day lags).
  • Days. Every date falls on a whole working day.
  • Minutes. Dates are exact to the working minute, and durations can be fractional days.

An explicit choice always wins and is kept across edits.

What Auto treats as sub-day information

  • Actual, constraint or planned times that fall inside a shift rather than on the calendar’s day boundaries.
  • Hour-based durations, or lags that aren’t whole days.
  • A data date with a time of day.
  • A part-day calendar: hours per day different from the shift length, part-day weekdays, or working exceptions shorter than a day.
  • The remaining completed-date convention, which needs minutes.

A midnight stamp, a stamp at the calendar’s first shift start or last shift end, and an end-of-day finish constraint don’t count. Auto stays in days if a calendar in use has no shift times, because minute precision needs them.

Why it matters

On the product’s corpus of 39 real P6 projects (26,337 compared values), whole days agree with P6 on 81.5% of values. Auto and forced minutes both agree on 87.6% (23,083 values). Auto never does worse than days on any project, and an automated test enforces that. Auto also detects not-started start milestones with no unfinished successor, whose late dates P6 stores at the end of a working day, so it now matches forced minutes on every project in the corpus.

How the choice survives export

P6 has no precision setting. When a plan is set to Days or Minutes, the XER or P6 XML export adds a text project UDF, Time precision (days or minutes), which P6 keeps. Re-importing the file restores your choice instead of auto-detecting again. An Auto plan writes nothing and is detected again on import, like any P6 file.

Long analyses run as background jobs

Large Monte Carlo runs, forensic windows, scenario calculations, schedule compression, resource leveling and program calculations can take longer than a web request should. These run as background analysis jobs on their own engine lane, so they don’t hold up the rest of the app.

  1. Start the analysis as usual. Small runs finish inline. A risk run queues as a job once activities × iterations passes 500,000. Other heavy analyses queue when they’re estimated to take more than about 10 seconds.
  2. Watch progress. The page shows the job’s status (queued, running, completed, failed or cancelled) and a progress percentage. You can leave and come back.
  3. Cancel if you need to. Cancelling stops the engine run straight away.
  4. Read the result. It’s saved like any other analysis, against the version it ran on. If the plan changed before the job started, the job fails rather than analysing the wrong version.

A job gets twice its estimated time, at least 2 minutes and at most 2 hours by default. A request estimated above that ceiling is refused with a prompt to reduce the iterations, inputs or steps. If you start the same request twice, you get the job already running rather than a second copy. Jobs need Schedule admin. Through the API, see analysis jobs.

  • Mistake: Forcing Days on an hourly P6 schedule to make it simpler

    In the app: Leave Auto on. Whole days round hour-based work and can move dates and float by a day.

  • Mistake: Re-importing the same 25 MB XER every month

    In the app: Import once and run the update cycle in the Planner. Re-import only when the source of truth is still P6.

  • Mistake: Expecting the retained P6 metadata on a 40 MB plan

    In the app: Check the import issues. If retention was dropped, the export carries only the modelled fields.

Keep exploring

Related across os.construction

Run it on your own project.

Start a 14-day free trial. Want a say in what ships next? Join the founding program.

Your role (optional)

14-day free trial. Pricing for paid plans isn't set yet. Talk to us. Already have an account? Sign in →

Ask the OS

Reading this page

Ask anything about this page. Answers use what you're reading as context.

Suggested

    AI · may be imperfect · sample data where shown

    Open file