LowestCode
Green AI Office and Business Application
by
One AI system, designed to replace
Microsoft Office, Microsoft Dynamics, Adobe, SAP and more.
One AI Workspace for Office & Business
With AI App Development Built In
In production since 2024 at an international insurance group: certified, due-diligence approved, and backed by the labor union. In the Quest, October to December 2026, one AI agent guided by one human creates an Office and Business App, with monthly, public, verifiable proof.
-
LowestCode at a glance: twenty pictures in one
-
LowestCode Unified Development System:
Coding, project management, testing 100 % by AI & LCP -
Energy and server hardware per 1,000 users
-
The workplace: several windows of one system side by side, in three themes
-
LowestCode Office-Text at work: one document in three windows, each scrolled on its own, built for full RPA
-
LowestCode Office-Sheet: a sheet with formulas and a chart inside the document, built for full RPA
-
LowestCode Office tool sets: the tools of the HTML editor, active while a text is edited
-
NoCode, LowCode and AI: the whole system behind every application
-
The Quest – updated every month
-
Import and export with Office tools
-
PMS: projects and workflows, declared and checked
-
PMS at work: a demo project on the board, with its sprint and timeline
-
The rollout and contribution chain
-
ERP-bound web shops, created in two days with AI chat and load-tested.
-
Sim-Test: synthetic data for development, test and trial
-
Innovation and cooperation
LowestCode AI brings Office, ERP, e-commerce, and software development into one workspace. Individuals and businesses get work done and create or extend applications as needed through AI chat.
All LowestCode office and business applications are lightweight, clean, and fast—secure, privacy-focused, and multilingual by design, with 21 languages from the start. They run in European cloud environments or on desktops and are designed to support compliance with EU requirements for data protection, cybersecurity, and AI.
Our objective is a fully automatable system with a clear, well-structured architecture, enabling end-to-end automation of office work, business processes, and application development.
No more manual coding for professionals and users: coding, project management, and testing run entirely through AI chat under LowestCode’s controls, supported by a transparent architecture made for AI.
A green revolution for business software: on average, 95 % less energy use and server hardware than six leading business systems, based on our own load measurements and estimated sizing for those six systems.
Build together, roll out downwards, contribute upwards: system developers, IT service providers, customer organizations, and users develop independently without interfering with one another, supported by automated rollouts and controlled contributions.
LowestCode is designed to replace applications such as Microsoft Office, Microsoft Dynamics, SAP, Adobe, Salesforce, and Mendix, bringing office work, business operations, design, and application development into one integrated system.
Users adapt and optimize their workflows and applications by talking with the integrated AI agent, which respects the permissions for documents, forms, and users.
All product names are trademarks of their respective owners.
One system in place of many applications What LowestCode is designed to replace, and where each part stands today.
Companies run their work on many separate applications: an office package, an ERP, a CRM, design tools and a platform to build their own. LowestCode is designed to replace them with one system, and to go beyond them: built new, without the legacy of decades, with a modern workplace of several windows side by side, faster work in the cloud and professional options from the start. The map shows where each part stands today.
The whole range of LowestCode
- DMS
- CRM
- HRM
- FIN
- PMS
- ERP
- E-Commerce
- Logistics
- Production
- Planning
- Insurance
- Office
- Website
- Design
- Communication
- Statistics
- Sim-Test
- Gaming
- Learning
- Java-App
- Python-App
- Desktop-App
- Mobile-App
- Native-Apps
- Data-Warehouse
- LCP
- MCP adapter
The color of a pill shows the state of its project: in production, in test or in preparation; a pill without color is planned.
✅ In production · 🟨 In test · 🟧 In preparation · ⬜ Planned · As of
What it is designed to replace
The percentage is our estimate of how much of each project's planned scope is built, not a comparison with the named product.
Designed to replace | LowestCode part | State · planned scope built | Effort so far |
|---|---|---|---|
Microsoft Office Office work | Office-Text Office-Sheet Office-Table Office-Layout Office Design | ✅ 95 % ✅ 90 % 🟨 85 % ✅ 90 % ✅ 85 % | 28 days 12 days 2 days 4 days 4 days |
Adobe Design, graphics, PDF | Design, Draw, Paint and PDF tools | ✅ | – |
Microsoft Dynamics, SAP ERP | ERP FIN ERP Web Shop | 🟨 30 % 🟨 30 % ✅ 60 % | 2 days < 1 day 4 days |
Salesforce CRM | CRM | Planned | – |
Mendix Application building | LowestCode LCP-Chat LCP * MCP adapter * | ✅ 80 % ✅ 50 % ✅ 45 % 🟧 | 6 years 6 days 2 days – |
And more Projects, documents, test data | PMS DMS Sim-Test | ✅ 50 % ✅ 75 % ✅ 55 % | 4 days 1 year 6 days |
State and effort come from the project table of the Quest. Effort is the running time of the work, the time elapsed, rounded; 10 hours count as one day. A dash marks a part without a value of its own; the tools of Office are described under Office tool sets. * LCP is in production; the MCP adapter, the adapter for the external Model Context Protocol, is in preparation.
All product names are trademarks of their respective owners.
An application, not a suite
A suite bundles separate products, a word processor beside a spreadsheet, an ERP beside a CRM, each with its own data, rights and screens, joined by interfaces. LowestCode is one application: documents, sheets, business data, rights and versions live in one shared model, so a new part extends the application instead of adding a product to connect.
The Quest: public proof of the system and of high-speed development
October to December 2026: one AI agent, guided by one human, creates the Office and Business Application, and the state and the effort of every project are published month by month. The map and the pills take their values from it.
The AI agent codes. LowestCode controls. 100 % AI chat execution in one LowestCode system: coding, project management, and testing. Simultaneously, controlled and safe.
An automation engineer's approach, not an IT one: every operation is a defined step, checked before it acts and measured after it, as in a plant. Data, code and user interface form one model, every change runs with its rights decided first, and its result is measured outside the agent. That is why the agent can code, structure and test the whole system safely, and why every step of the system can be automated.
LowestCode is what AI needs: NoCode and LowCode in one integral, compact system for Office and business applications, and LCP controls every step. The result is controlled agent coding, high coding efficiency, and a clear interface for people.
95 % less energy and server hardware on average Documented resource efficiency: own load measurements and estimated sizing for six leading business systems; SAP comes closest with in-memory.
LowestCode keeps about one core busy for 1,000 users, measured; sized with headroom and a standby, that is four cores, eight with high availability on two sites. Five of the six other systems need far more, estimated from their published sizing rules, benchmarks and typical landscapes.
- 1.4 billionrows read by the database in one run
- 30 minutesof full load
- 171,000checks of the answers, none failed
- 9 to 136 msstandard response time: 50 % to 95 % of the requests
- ≈ 5 wattsprocessor power, load generator included
The web shops were measured the same way; their figures stand in the web shop picture and the load test report.
The numbers, their sources and the measurement are on these pages; no competitor product was measured. All product names are trademarks of their respective owners.
Power users deliver.
IT experts guide.
Development partners join through one governed rollout chain.
The LowestCode rollout system lets development partners build independently without interfering with one another, as the integrated AI agent does the coding and keeps it consistent.
- System
- Professionals
- Organization*
- User
We roll out code and data: documents and information such as taxes and legal information, where each customer can choose which services to implement.
Contributions travel up the chain in a controlled way. Once we are established on the market, we plan a participation system that rewards lasting contributions.
- User
- Organization*
- Professionals
- System
In addition, every partner installation rolls out from local development to production.
* In LowestCode development, the Organization level is played by the insurance use case, in practice since 2024.
LowestCode Office at work Text documents, tables, presentations, notes, reports and websites are forms of one governed document, not separate programs.
LowestCode Office is the working surface of the system, with the HTML editor and the form generator at its core: documents are written in the editor, forms are generated by the system rather than programmed, and both keep their rights, versions and languages in the system.
- Text
- Design & Publish
- Markdown
- Table
- Grid
- Sheet
- Project
- Figure
- Media
- Draw
- Paint
- Chart
- Layer
- Layout
- Version
- Binder
- Languages
- Properties
- Forms
- Secure Document AI Chat Agent
- Themes & Formats
Import and export: Word, Excel, PowerPoint and PDF are formats LowestCode Office imports and exports with its own means; the document itself, with its rights, versions and languages, stays in the system. All product names are trademarks of their respective owners.
Data protection and security by design
A LowestCode document is not a file on a disk or a share. It lives in the system, so the same rights, owners and versions decide who reads and changes it, people and the AI agent alike. A Word, Excel or PDF file is an export, not the document. Data protection and security come from how Office is built, not from a feature added afterwards.
One workplace.
One development IDE.
Many windows.
One shell for any application, LowestCode Office included.
Your windows, ready when you log in
From a single window to a complex working and development environment, the workplace fits the task. Your windows and their layout are saved and preloaded, ready the moment you log in. Working environments can be defined for each task and guarded by rights. Windows sit side by side, and the same document can be open in many of them, working in sync. The same system is your workplace and your development and test environment, so you build your own applications there and test them on synthetic data before they reach production. The test environment is independent of production, separate in data and files, and always available and accessible.
One menu, two views, one toolbar
The navigation is one menu shown two ways, as a side tree or as a top menu. The side menu doubles as the toolbar, with the HTML editor's controls in it, as on the left of the picture. Themes and formats change the look, not the content.
The development IDE inside the workplace
The development IDE is not a separate program. It uses the same multi-window workplace: the code walkthrough opens beside the forms and documents of the day, visible to developers by rights and invisible to everyone else. Users work in the normal workplace and never meet the coding capability.
An ERP for industry, from the order to the machine Trade, production, machine building and logistics in one system, built on a real industrial business.
We build the ERP on a real business: a metal profiling company that trades and sells online, profiles steel from the coil, sends profiles out for coating, builds machines for itself and for its customers, maintains them, takes on robot automation projects, and runs its own logistics and transport. The web shops are part of the ERP: their pages show the products, prices and taxes the back office maintains, as they stand at the moment of the view, and every order lands with its lines, addresses and payment where the back office works with it.
Sim-Test: synthetic data for every proof What a company gets from it: data checks, security, load checks and planning simulations.
Every company needs this, and most ignore it: software has to be proven before real data and real money depend on it, and proven without handing customers' data to developers, testers or an AI. Sim-Test makes both possible: synthetic data in the shape of your business, generated from your own structures.
A structured innovation portfolio grounded in working software Deterministic AI, internationalization, digital sovereignty, and measurable efficiency.
We have structured a portfolio of 94 innovations and innovation candidates, including 43 system-level architecture innovations (A-category). They rethink how database semantics, documents and interfaces, automation, AI governance, security, and deployment work together. On this basis we are preparing a research grant application to the German research ministry (BMFTR, formerly BMBF) and are looking for a university partner.
For customers: What we modernize.
The business
Model: we deliver the technology, and partners bring it to customers. Revenue comes from licenses per user and month, through partners and direct, and from building applications for customers with the AI agent.
First market: mid-sized companies, insurers and organizations in the German-speaking countries that modernize Office work, processes, Excel and VBA. We enter through concrete projects.
Market: Office and ERP are the core markets; Gartner forecasts USD 58.2 billion for low-code technologies worldwide in 2029 (forecast analysis of November 2025).
Expansion: partners offer LowestCode as their own application layer; AI providers and cloud hosts add it to their models and hosting.
Investment: we are looking for strategic investors who are partners as well, and for business angels. The funds go mainly into marketing and sales, then into partner engineering and operations, toward one goal: LowestCode in common use, nationally and across Europe.
LowestCode in short One integral, compact system for Office and business applications, and LCP controls every step. Many problems solved.
In use and measured today
- ✅Ten years of development, proven in real use: insurance software certified and in production since 2024 at an international insurance group, with due diligence passed and labor union approval.
- 🧠We don't write code anymore: since September 2026, coding, project management and testing run 100 % by AI chat, on the NoCode & LowCode (NCLC) system with the LowestCode Communication Protocol (LCP), which checks each request before it runs.
- ⚡About one busy core for 1,000 users, measured: in one 44-minute load run, 30 minutes of it at full load, the database reads more than 1.4 billion rows, and the device does not even get warm.
- ⚡Green tech: 95 % less energy and server hardware on average than six leading business systems, from our own load measurements and estimated sizing for the six; SAP comes closest with in-memory.
- 🌍21 application languages from the start, and each partner can add languages or define dialect variants; right-to-left scripts, time zones and country formats are built in.
- 🆓No license fees for components: every third-party component is free open source. AI agents and cloud hosting are external services with their own terms.
How it works
- 📄Office, business applications and ERP in one workspace: LowestCode Office stands built since 8 November 2026 and is in our own use, and the ERP with its web shops is being built in the same system.
- 🏅One rollout and contribution chain: System ⬌ Professionals ⬌ Organization ⬌ User. Partners build independently, and contributions travel back up in a controlled way.
- ⚙️An automation engineer's approach toward a fully automatable system: every operation is a defined step, checked before it acts and measured after it, as in a plant, and people, workflows and the AI agent run the same steps.
- 🔒Data protection and security by design: every document and record keeps its rights, owners and versions in the system, for people and the AI agent alike; development and test run on synthetic data. Designed for the EU rules: GDPR and EU cybersecurity for the cloud, the EU AI Act for the AI.
We set the direction. AI agents code. LCP controls.
© 1995-2026 Dr.-Ing. Raimund Reisacher –
LowestCode technology
German patent application DE 10 2021 103 571 –
relational database system, pending
© 1995-2026 rsaccess GmbH –
LowestCode website, marketing and sales
LowestCode © Dr.-Ing. Raimund Reisacher · hosted by rsaccess GmbH
This document is available in English only. It is the text we check and verify.
LowestCode Load Test Report
This report shows how we measured the server load of LowestCode: what ran, on which machine, what was counted, and what came out. It belongs to the energy and server comparison. Since October it also covers the web shop of LowestCode, measured the same way. State: 11 October 2026.
The result in short
- 1,000 simulated office users worked on LowestCode for 30 minutes at full load, one action every 15 seconds each.
- Server and database together kept 1.2 processor cores busy. They ran on one device, an Apple M1 Pro, which also ran the load generator.
- In one run the users performed about 141,000 actions with 172,000 requests. Every answer was checked for its content, about 171,000 checks, and in the three runs after we corrected the test data none failed.
- Half of all requests were answered within 9 milliseconds, 95 % within 136 milliseconds.
- The processor drew about 5 watts for the whole test, load generator included. The device did not get warm.
- The web shop, measured the same way: 1,000 simulated shoppers kept 0.67 processor cores busy, and 95 % of the requests were answered within 24 milliseconds. 5,000 shoppers ran on the same device too; once we had fixed what the test found, not one of their 1.2 million requests failed. The details are in the section The web shop.
Five office runs count. A sixth was aborted and is not counted.
What was tested
The test runs the office work of our insurance application in a test environment with synthetic data: 1,000 users, 16,000 customers and 16,000 offers. No customer data is involved.
Each simulated user does what an office user does. The user logs in once, which loads the workplace page and the offer list. Then the user repeats: think for about 15 seconds, then do one of seven things.
|
Action
|
Share
|
What happens
|
|---|---|---|
|
Open a child table
|
30 %
|
the final texts of one offer are read
|
|
Filter a list
|
25 %
|
the offer list is filtered by a product
|
|
Open an offer
|
15 %
|
the offer and its details are read, two requests
|
|
Save a row
|
15 %
|
one field of a customer is changed and stored
|
|
Search
|
10 %
|
a customer is searched by name
|
|
Calculate a tariff
|
4 %
|
the tariff document of an offer is calculated and rendered, two requests
|
|
Print an offer letter
|
1 %
|
the offer letter is rendered
|
The think time varies between 10.5 and 19.5 seconds. The choice of action and offer comes from a generator with a fixed seed, so two runs replay the same sequence.
The machine and the setup
- One device, an Apple M1 Pro with 10 cores and 32 GB of memory.
- On it: the LowestCode server, the PostgreSQL 14 database with its default settings (128 MB shared buffers), and the load generator k6.
- The database held 811 MB.
The load generator ran on the same machine, so its work is part of the measured machine total. A server in a data center was not part of this test.
How one run goes
One run takes 44 minutes.
|
Phase
|
Duration
|
What happens
|
|---|---|---|
|
Idle
|
3 minutes
|
the server runs, nothing loads it; this is the baseline
|
|
Ramp-up
|
10 minutes
|
the users log in one after another until 1,000 are working
|
|
Full load
|
30 minutes
|
1,000 users work; all load and power figures come from this window
|
|
Ramp-down
|
up to 1 minute
|
the users stop
|
What we count
Three words are easy to mix up, so here is what each means.
- An action is one thing a user does: open an offer, save a row, calculate a tariff. About 141,000 per run, 58 per second.
- A request is one call to the server. Most actions send one, two of them send two, and the login sends four. About 172,000 per run, 70 per second.
- A check is our test of one answer. The answer must arrive without error and contain what was asked for: the rows of the requested offer, the saved row, the right customer in the tariff document. An answer that arrives but carries the wrong content counts as a failure. Every request is checked except the first load of the public page, about 171,000 checks per run.
Rows read by the database are a fourth number, and a much larger one. It is not a count of operations; it is explained below.
The five runs
|
Run
|
State of the system
|
Actions
|
Requests
|
Failed requests
|
Checks
|
Failed checks
|
|---|---|---|---|---|---|---|
|
1
|
before an index rule
|
141,043
|
172,050
|
0
|
171,050
|
97
|
|
2
|
same as run 1
|
141,017
|
172,013
|
0
|
171,013
|
38
|
|
3
|
index rule in place
|
141,175
|
172,206
|
0
|
171,206
|
0
|
|
4
|
one more query improved
|
141,184
|
172,216
|
0
|
171,216
|
0
|
|
5
|
as run 4, with a profiler attached
|
141,162
|
172,187
|
0
|
171,187
|
0
|
The failed checks of runs 1 and 2 were all tariff documents that lacked the customer reference our check looks for. The cause was a defect in the generated test data, not in the application: 149 of the 16,000 test customers carried a search key in a form the application itself never writes. We corrected the test data generator before run 3. No request failed in any run, and the server logged no error.
Runs 1 and 2 were made at the same state to see how well a run repeats: every figure agreed within 2 %. Runs 3 and 4 repeat within 1 % on the measured processes.
Response times
Over all requests of a run:
|
Run
|
Half of the requests within
|
95 % within
|
99 % within
|
|---|---|---|---|
|
1
|
18 ms
|
208 ms
|
374 ms
|
|
2
|
19 ms
|
209 ms
|
381 ms
|
|
3
|
9 ms
|
136 ms
|
289 ms
|
|
4
|
9 ms
|
137 ms
|
288 ms
|
|
5
|
9 ms
|
147 ms
|
304 ms
|
Per action, run 4:
|
Action
|
Count
|
Half within
|
95 % within
|
|---|---|---|---|
|
Open a child table
|
42,115
|
4 ms
|
39 ms
|
|
Filter a list
|
35,176
|
28 ms
|
63 ms
|
|
Open an offer
|
21,393
|
7 ms
|
53 ms
|
|
Save a row
|
21,303
|
10 ms
|
46 ms
|
|
Search
|
14,145
|
20 ms
|
61 ms
|
|
Calculate a tariff
|
5,639
|
182 ms
|
299 ms
|
|
Print an offer letter
|
1,413
|
78 ms
|
116 ms
|
|
Login
|
1,000
|
228 ms
|
271 ms
|
|
Load the workplace page
|
1,000
|
449 ms
|
484 ms
|
The workplace page is the whole workplace with its preloaded windows, about 8.5 MB. The tariff document is about 3.2 MB.
Our own target is 95 % of all requests within 100 milliseconds. At this load on this device we measured 136 milliseconds, so the target is not met yet.
On 10 October we ran the office test once more on the code of that day, as run 6: 1.17 processor cores busy, half of the requests within 9 milliseconds, 95 % within 117 milliseconds. Closer, still not met.
Processor load and power
The figures are the averages of the 30 minutes at full load, sampled every 5 seconds.
|
Run
|
Server
|
Database
|
Server and database
|
Processor power, whole machine
|
|---|---|---|---|---|
|
1
|
0.52 cores
|
1.21 cores
|
1.73 cores
|
6.8 W
|
|
2
|
0.53 cores
|
1.23 cores
|
1.76 cores
|
6.9 W
|
|
3
|
0.54 cores
|
0.67 cores
|
1.21 cores
|
4.5 W
|
|
4
|
0.54 cores
|
0.66 cores
|
1.20 cores
|
5.2 W
|
- The processor power is that of the whole machine: server, database, load generator and the operating system together. The share of server and database is 2.2 to 2.7 watts.
- It is the power of the processor as the operating system reports it. It is not a measurement at the wall socket.
- Run 4 started at midnight beside the device's nightly maintenance jobs, so its machine total is higher than in run 3. The server and the database measure the same.
- Run 5 ran with a profiler attached to the server. The profiler costs processor time of its own, so the load of that run is not comparable, and no power was sampled.
That the device did not get warm is an observation of the person at the desk, and it fits the 5 watts.
What the database did
Per run the database processed about 2.5 million transactions. To answer them it read rows: 1.44 billion by scanning tables and another 310 to 340 million through an index.
This number is large because it counts every row the database touches, not every answer it gives. A filter over a list of 16,000 offers touches many thousands of rows to return one page of them.
It also shows what the test is good for. In runs 1 and 2 the database read 5.45 billion rows per run. The statistics showed where: columns that link a child to its parent had no index. We made the index a rule of the structure generator, and from run 3 on the database read 1.44 billion rows and needed 0.67 instead of 1.22 cores.
The web shop
LowestCode has a web shop inside its ERP since October 2026: product list, product page, cart, checkout and order, on the products, prices and taxes of the back office. We load-tested it the way we tested the office work: the same device, the same 15 seconds of thinking between two actions, 30 minutes of full load, every answer checked for its content.
What the shoppers do
Each simulated shopper loads the home page, logs in and opens the shop. Then the shopper repeats: think for about 15 seconds, then do one thing.
|
Action
|
Share
|
What happens
|
|---|---|---|
|
Look at a product
|
30 %
|
the product page with all its pictures
|
|
Browse the list
|
30 %
|
the first page of the list, a category, or the next page, 10 % each
|
|
Add to the cart
|
15 %
|
one product goes into the cart
|
|
Open the cart
|
8 %
|
the cart page with its lines
|
|
Search
|
5 %
|
a product is searched by name
|
|
Change the cart
|
5 %
|
a line is changed or removed
|
|
Order
|
5 %
|
the checkout page, then the order is placed
|
|
Pay
|
2 %
|
an open order is paid
|
The shop holds 1,000 products in 20 categories, made from our 50 sample products, each with its pictures. The payment is simulated: no money moves and no payment provider is involved.
1,000 shoppers
|
Measured
|
Run of 9 October
|
Run of 10 October
|
|---|---|---|
|
What the shop was
|
one list of all 1,000 products, no pictures
|
the shop as it is today: 20 products per page, every picture loaded
|
|
Requests
|
158,695
|
240,520
|
|
Checks of the answers
|
about 158,000, none failed
|
239,520, none failed
|
|
Half of the requests within
|
12 ms
|
8 ms
|
|
95 % within
|
34 ms
|
24 ms
|
|
Processor cores busy, server and database
|
0.83
|
0.67
|
The two runs measure two different shops, so both stand. The run of 10 October is the one that counts for the shop of today. It loaded 81,237 pictures, 95 % of them within 7 milliseconds, and its shoppers placed about 8,000 orders, each checked afterwards for its lines and its payment. The target of 95 % within 100 milliseconds is met at this load.
The processor power of the machine was about 3.6 watts. The load generator, which parses every page the shoppers see, takes almost as many cores as server and database together, so the power of the machine is not the shop's own and we derive no share from it.
5,000 shoppers
On 10 October we ran the shop with 5,000 shoppers on the same device. Five times the load is where weak spots show, and it showed three:
- The connection pool of the server locked up during the ramp-up: one route held its database connection while it waited. We fixed the route and repeated the run.
- The cart is found by its customer, and that column had no index. It took 68 % of the database time. With the index the database needed 1.22 instead of 2.64 cores.
- A server running on a developer's machine reloads part of its cache at every login. A cloud server never does that. Switched off, the slowest answers came down from seconds to a fifth of a second.
The last of these runs: 5,000 shoppers, 1.2 million requests, none failed. Server and database kept 2.62 cores busy, half of the requests were answered within 8 milliseconds, 95 % within 141 and 99 % within 222 milliseconds. The device ran at about 7 of its 10 cores, 3 of them for the load generator. A further run with twice the connections measured the same, so the connection pool is not what keeps the target of 100 milliseconds out of reach here. Our reading is the shared device, and we cannot test it without a second machine.
What it means for a server
Sized by the same rule as the office test, with headroom, a standby and, for high availability, a second site: 2.4 to 4.8 server cores and about 10.5 to 17 watts per 1,000 shoppers. 1,000 and 5,000 shoppers each run on one server with 8 cores plus a standby. This is calculated from the measurement, not measured on a server.
What is measured and what is not
Measured, on this device:
- the load of 1,000 office users, and of 1,000 and 5,000 shoppers in the web shop, with their actions, requests and checked answers
- the response times
- the processor load of server and database, and the processor power of the machine
Calculated from the measurement, not measured:
- the server sizes of the comparison on our homepage, 4 to 8 cores and 20 to 31 watts per 1,000 users. They size the measured load for real operation, with headroom, a standby and a second site, on the published power data of server machines.
- the server sizes of the web shop, 2.4 to 4.8 cores and about 10.5 to 17 watts per 1,000 shoppers, by the same rule.
Not measured at all:
- any other product. The six systems of the comparison are estimated from their own published sizing rules and certified benchmarks. No other web shop was measured.
- power at the wall socket, and a server in a data center
- more than 1,000 office users or 5,000 shoppers on one machine, and the memory they would need
- a payment provider and the stock booking of an order; the shop's payment is simulated
- the message queue service, which did not run in this test
The test data is synthetic. It proves how the system behaves, not how real customer data is distributed.
The evidence
Every run, of the office test and of the web shop, left one set of files: the setup with the software revision, the result of the load generator, the processor and power samples, and the statistics of the database. The files are kept unchanged, with a checksum for each, as a versioned evidence package. The test instrument is part of the source code, so a run can be repeated.
We open the package in a technical review.