Cloud vs on-prem deployment for advanced planning: performance trade-offs and integration challenges

MES GE Vernova
, , , , , , , ,
1

Our organization is evaluating cloud migration for Ge Proficy Smart Factory 2021.2 advanced planning module. Currently running on-premises with dedicated servers, and we’re seeing pressure to move to Azure cloud for “better scalability and cost savings.” I’m skeptical about the performance claims, especially for our complex planning scenarios with 200+ work centers and 5000+ active orders.

Our on-prem setup handles planning runs in 8-12 minutes with predictable performance. Finance is pushing cloud based on vendor promises of elastic scaling, but I’m concerned about latency, integration with our legacy ERP system (which must stay on-prem for now), and whether cloud-native really delivers for manufacturing planning workloads. We’ve been running this system reliably for three years and I don’t want to create problems chasing theoretical benefits.

Would appreciate hearing from anyone who’s made this transition - what were the actual performance differences, how did you handle legacy integration challenges, and did you consider hybrid deployment options as a middle ground?

2

Let me provide a comprehensive analysis of all three key considerations: elastic scaling in cloud, legacy integration challenges, and hybrid deployment options.

Elastic Scaling in Cloud - Reality Check:

The elastic scaling promise is real but nuanced for manufacturing planning. Unlike web applications with unpredictable traffic spikes, your planning workload is highly predictable. You run planning 2-4 times per day at scheduled intervals. Traditional elastic scaling doesn’t help much here.

However, cloud does offer value through “scheduled scaling.” Configure your planning VMs to scale up 30 minutes before planning runs, then scale down afterward. Example: Run 4 vCPU instances normally ($200/month), scale to 32 vCPU for planning windows ($1600/month if running 24/7, but only $80/month for 2 hours daily). This is where cloud economics actually work for manufacturing.

Performance comparison from our deployments:

  • On-prem (16-core server): 10 minute planning runs
  • Cloud default (8 vCPU): 18 minute runs (worse)
  • Cloud optimized (16 vCPU, premium SSD): 7 minute runs (better)
  • Cloud scaled (32 vCPU during planning): 4 minute runs (significantly better)

The key is proper architecture, not just lifting-and-shifting to cloud.

Legacy Integration Challenges - Three Approaches:

  1. Direct VPN Integration (Simplest): Site-to-site VPN between on-prem and cloud. Works for moderate data volumes (<1GB daily). Latency typically 40-80ms. Cost: $50-200/month. Good enough if your ERP integration is already batch-oriented.

  2. ExpressRoute/Direct Connect (Best Performance): Dedicated fiber between your datacenter and cloud provider. Latency <10ms. Cost: $5,000-8,000/month. Only justified for high-volume real-time integration or if you’re moving multiple systems to cloud.

  3. Data Synchronization Layer (Recommended): Implement staging database in cloud that syncs from on-prem ERP every 10-30 minutes. Smart Factory reads from cloud staging, eliminating cross-network calls during planning. Use Azure Data Factory or AWS Glue for sync orchestration. Cost: $200-400/month. Requires process adaptation but solves 90% of integration problems.

Your real-time order updates concern: 15-minute sync intervals are acceptable for most manufacturing. Truly urgent orders can be manually expedited. If you genuinely need sub-minute integration, you need ExpressRoute or should stay on-prem.

Hybrid Deployment Options - Four Models:

  1. Planning On-Prem, Analytics Cloud: Keep advanced planning module on-prem near ERP. Move reporting, analytics, and dashboards to cloud. Best of both worlds - no integration latency, but modern cloud analytics. This is my recommendation for your situation.

  2. Full Cloud, Sync Integration: Move everything to cloud, implement staging database sync from ERP. Works well if you can tolerate 15-30 minute data lag and are willing to re-architect integration.

  3. Edge Computing Model: Run lightweight planning engine on-prem for immediate scheduling, sync to cloud for global optimization and analytics. Complex but powerful for multi-site operations.

  4. Phased Migration: Start with non-critical modules in cloud (quality management, reporting), keep planning on-prem. Migrate planning last after proving cloud architecture. Reduces risk.

My Recommendation for Your Scenario:

Given your concerns about legacy ERP integration and stable on-prem performance, implement Hybrid Model #1: Keep advanced planning on-premises, move reporting and analytics to cloud. This gives you:

  • No integration latency issues (planning stays near ERP)
  • Cloud benefits for analytics and dashboards
  • Lower migration risk
  • Proof of concept for broader cloud adoption
  • Cost savings on reporting infrastructure

If finance insists on full cloud migration, demand proper TCO analysis including:

  • ExpressRoute or equivalent networking ($60K-100K annually)
  • Re-architecture effort (6-12 months, $200K-500K)
  • Right-sized compute instances (not default sizing)
  • Data egress costs (often overlooked, can be $500-2000/month)

Cloud can work for manufacturing planning, but it requires proper architecture and realistic cost analysis. Don’t let vendor promises or finance pressure drive poor technical decisions. Your stable on-prem system has real value.

3

Don’t let finance drive a technical decision. Cloud isn’t always cheaper - we ran TCO analysis and found cloud costs 40% more over 5 years when you factor in proper networking, security, and the re-architecture work. The elastic scaling in cloud sounds great in theory but manufacturing planning isn’t like web traffic. Your planning load is predictable - you don’t need elastic scaling, you need consistent performance. If your on-prem system works, keep it until you have a real business reason to change.

4

That’s exactly what I’m worried about. Our current integration does real-time order updates from ERP. Moving to batch would require significant process changes and might impact our ability to respond to urgent orders. Did you lose any planning flexibility with the batch approach?

5

The legacy ERP integration is your real challenge, not performance. We kept our ERP on-prem and moved Smart Factory to Azure. The network latency between cloud and on-prem for data sync killed us initially. Had to implement Azure ExpressRoute (expensive) and redesign our integration to use batch transfers instead of real-time sync. Went from 200+ API calls per hour to 6 scheduled batch jobs per day. Changed our planning model but it works.

6

Consider a hybrid deployment. Keep your planning engine on-prem where it can access ERP quickly, but move reporting and analytics to cloud. We did this with Smart Factory 2022.1 - planning module stays local, but we use Azure for data warehousing and Power BI dashboards. Get the cloud benefits for analytics without the integration headaches. The hybrid deployment options in newer versions support this architecture pretty well.