AADC Deployment Strategies - On-Premises, Cloud and Hybrid

Binary Kernels: The Key to Deployment Flexibility

When AADC compiles your pricing models it produces binary kernels: machine code that runs 6 to 1000× faster than the original.

Serialisable

save to disk, transfer anywhere

Language-agnostic

call from Python, C++, C#, or Excel

No source code

only binary machine code, not source

Thread-safe

automatic parallelisation built-in

This gives you deployment flexibility that traditional code doesn't offer. Source code stays on-premises; only binary kernels move to the cloud.

In-House Deployment

Run AADC on your own infrastructure - from individual machines to enterprise grids

AADC binary kernels can be deployed entirely on-premises, giving you complete control over your infrastructure and data.

Complete Control

Full ownership of infrastructure and data

✓ Data never leaves your network
✓ Complete control over hardware selection
✓ No external dependencies
✓ Compliance-friendly for regulated industries
✓ Integrate with existing security policies

Individual Machine Deployment

Run on developer workstations or single servers

✓ No infrastructure setup required
✓ Ideal for development and testing
✓ Full AAD capabilities on a laptop
✓ AVX2/AVX512 optimisation on modern CPUs
✓ Same code runs in production

Enterprise Grid Deployment

Scale across your data center infrastructure

✓ Linear scaling across compute nodes
✓ Integration with job schedulers (LSF, PBS, SLURM)
✓ Shared kernel cache across grid
✓ High-availability configurations
✓ Fits existing enterprise architecture

System Requirements

CPU: Modern x86-64 with AVX2 (AVX512 recommended)

Memory: Scales with model complexity (typically 4-32GB per core)

OS: Linux (RHEL, Ubuntu, CentOS), Windows Server

Integration: C++, Python, C# APIs available

Cloud Deployment

Secure, scalable cloud execution with up to 99% cost reduction

With MatLogica's AADC, source code stays on-premises. You deploy only binary kernels - source code is not included.

Encrypted Cloud Deployment

Ship only encrypted binary computational graphs to the cloud

✓ Proprietary models stay on-premises
✓ Algorithms never exposed
✓ Sensitive data remains secure
✓ Source code not included in deployed kernels
✓ Compliance-friendly architecture

Up to 99% Cost Reduction

Lower cloud bills from faster execution

✓ Fewer instances needed
✓ Shorter runtime per calculation
✓ AVX2/AVX512 vectorisation
✓ Optimal multi-core utilisation
✓ Pay only for actual compute used

Accelerated Models with AAD

Speed up calculations while computing all sensitivities

✓ 20-50x faster execution
✓ Automatic sensitivity calculation
✓ All Greeks computed together
✓ Thread-safe by design
✓ Linear scaling to available cores

Proposed Deployment Architecture

On-premises security with cloud scale

In the hybrid model, MatLogica AADC is part of the Quant Library on-premises. When a request to price a trade comes in.

MatLogica Hybrid Deployment Architecture

On-Premises

✓ Proprietary models stay secure
✓ Sensitive portfolio data protected
✓ Kernel generation happens here
✓ Full control over IP

Cloud

✓ Encrypted binary kernels only
✓ Burst compute on demand
✓ Multi-region scaling
✓ Pay only for execution time

How Hybrid Deployment Works

Request Arrives

Pricing or risk request arrives at on-premises Quant Library

Kernel Check

System checks if kernel for this configuration exists

Generate or Retrieve

If missing, AADC generates kernel on-premises; otherwise retrieve cached kernel

Cloud Execution

Encrypted kernel executes on cloud infrastructure with market data

Results Returned

Results aggregated and returned to requesting system

Want to understand how to design a Live Risk Solution using Kernels?

Learn how kernels support daily trading workflows, from model calibration to sub-millisecond RFQ responses, and how modular design enables true Live Risk.

Live Risk Architecture Blueprint

You've integrated AADC and now you're designing a full live risk system. This optional blueprint collects patterns from working implementations; use what fits and ignore what doesn't.

Explore Live Risk Architecture

Hybrid Use Cases

Burst Capacity

Run normal workloads on-premises, burst to cloud for end-of-month or stress testing

Geographic Distribution

On-premises in headquarters, cloud execution in regional offices

Disaster Recovery

Primary on-premises with cloud-based failover capability

Development to Production

Develop on-premises, deploy production kernels to cloud

Frequently Asked Questions

Can AADC run on individual developer machines?

Yes, AADC binary kernels can run on any machine with a modern CPU supporting AVX2 instructions. Developers can work on their laptops or workstations with full AADC capabilities, then deploy the same code to production infrastructure. This enables a consistent development-to-production workflow.

What are the deployment options for AADC?

AADC supports three deployment models: (1) In-House - run entirely on your own infrastructure from individual machines to enterprise grids; (2) Cloud - deploy encrypted binary kernels to AWS, Azure, GCP, or any cloud provider with up to 99% cost savings; (3) Hybrid - keep models and sensitive data on-premises while bursting compute to the cloud. Choose based on your security, performance, and cost requirements.

How secure are AADC binary kernels when deployed to the cloud?

AADC binary kernels contain only machine code - source code is not included. Your source code and sensitive data remain on-premises while only binary kernels are deployed to the cloud.

How much can I reduce my cloud costs using MatLogica AADC?

You can reduce your cloud bill by up to 99% by introducing MatLogica and improving software scalability through efficient binary kernels, AVX2/AVX512 vectorisation, and optimal multi-core utilisation. A typical comparison shows ~$150/month with AADC kernels vs ~$15,000/month with traditional always-on approaches for the same workload.

Can I keep my data on-premises while using cloud compute?

Yes, with hybrid deployment you ship only a binary version of the computational graph to the cloud, leaving your data and analytics on-premises. This approach provides cloud scalability without exposing proprietary information, maintaining complete data sovereignty.

How does hybrid cloud deployment work with AADC?

With AADC hybrid cloud, you record and compile computational graphs on-premises where your proprietary models reside. Only the encrypted binary kernels are deployed to cloud infrastructure. Market data can be fed directly to cloud instances, and results returned, without exposing your algorithms or sensitive portfolio data.

What cloud providers are supported?

AADC binary kernels are cloud-agnostic and run on any cloud provider including AWS, Azure, Google Cloud, and private clouds like OpenStack. The kernels execute as native binaries, requiring only standard CPU compute instances with AVX2/AVX512 support for optimal performance.

What are the system requirements for on-premises deployment?

AADC requires a modern x86-64 CPU with AVX2 support (AVX512 recommended for optimal performance). Memory requirements scale with model complexity, typically 4-32GB per core. Supported operating systems include Linux (RHEL, Ubuntu, CentOS) and Windows Server. APIs are available for C++, Python, and C#.

Deploy AADC Your Way