How Salesforce, GitLab, and Atlassian Do On-Prem — and Why Our Model Is Different

Seven enterprise tools compared on their on-prem deployment models. What each *assumes*, who each *excludes*, and why CollabOps started from a *different assumption*.

John Baek
John Baek
Founder, CollabOps
How Salesforce, GitLab, and Atlassian Do On-Prem — and Why Our Model Is Different

A common opening line in customer meetings: "Do you deploy like GitLab Self-managed?"

Answer: no. Why is the content of this post. We break down seven enterprise tools' on-prem models, then look at what each assumes.

Seven-tool comparison

Tool                       | On-prem model               | Min requirement              | Assumption                      | Customers excluded
───────────────────────────┼─────────────────────────────┼──────────────────────────────┼────────────────────────────────┼───────────────────────
Salesforce                 | Hyperforce (SaaS in effect) | (none — SaaS)                | All data can live in cloud      | 90% of KR public/finance
GitLab Self-managed       | docker-compose / Helm       | 4 vCPU (5 users) → 50+ k8s   | Past 50 users, k8s available    | Mid-size, no-k8s shops
GitLab Dedicated           | AWS-managed onprem-like     | (managed)                    | AWS usage acceptable            | Air-gapped, non-AWS
Atlassian Data Center     | k8s + external DB           | k8s + Postgres + LB          | Infra expertise present         | Ops teams under 5 people
GitHub Enterprise (GHES)   | VMware OVA                   | 4 vCPU + VMware/AWS          | VMware or cloud                 | KVM, Hyper-V, OpenShift
Jenkins (LTS)              | Self-managed jar             | Java + DB                    | Self-operated                   | No DevOps headcount
Sentry self-hosted         | docker-compose               | Docker host                  | Not recommended for production (per docs) | Production usage
CollabOps                  | 5-class taxonomy (single VM · Helm · OVA · Operator · tarball) | 1 VM + 8 vCPU | No assumption of infra expertise | (deliberately almost none)

The implicit message of each minimum

A tool's minimum requirement isn't a technical spec. It's a customer-segment selection.

Salesforce's message

"We do SaaS only. If you need on-prem, look elsewhere." Hyperforce runs on hyperscalers in customer regions, but not in customer-owned datacenters.

Markets excluded — Korean public agencies, the financial sector (especially K-ISMS-bound), Japanese government, parts of EU regulation. Tens of billions of dollars of market deliberately excluded. That is Salesforce's strategy.

GitLab's message

"Under 50 users, run on 4 vCPU. Past 50, adopt k8s." This message excludes the market that is under 50 users + has no plans to adopt k8s. That is the core market of Korean and Japanese regional banks and insurers.

Atlassian's message

"You have an infra-capable IT team." k8s + external DB + LB + shared file system = expertise across four areas. For a 5-person IT shop, four areas is too much.

GHES's message

"VMware or AWS/GCP/Azure." No k8s — paradoxically, GHES does not install on k8s. Customers on KVM, Hyper-V, or OpenShift need separate wrapping.

CollabOps's different assumption

The six tools above all match their respective ICPs. CollabOps targets a different ICP with a different assumption:

Assumption                                                | Result
──────────────────────────────────────────────────────────┼────────────────────────────
No assumption of customer infra expertise                 | Five environment classes supported
Don't block customer's *next* decision                    | Embedded lightweight k8s → migration path to existing k8s
Don't infringe customer's data boundary                   | Air-gapped supported as default
SaaS is the *option*; self-hosted is the *default*        | → /blog/product/why-self-hosted-default

These four assumptions align every deploy artifact decision. Change one assumption, the whole stack takes a different shape.

Who should pick which tool

Objectively:

Condition                                          | Recommendation
───────────────────────────────────────────────────┼─────────────────────────
All data in cloud OK + standard ICP                | Salesforce / SaaS DevOps
50+ users + k8s ops team                           | GitLab / Atlassian
VMware standard + some cloud OK                    | GHES / GitLab Dedicated
Public sector / finance / defense + air-gapped or in-house | CollabOps
≤5 infra people + own datacenter                    | CollabOps

This isn't sales material. It's a decision matrix. CollabOps isn't the answer for every environment — our space is the environments other tools can't enter.

The real conclusion of seven comparisons

Each tool's minimum requirements reveal its origin.

  • Salesforce — cloud-first origin → on-prem options retrofitted → effectively SaaS
  • GitLab — open-source self-host origin → past 50 users assumed out of scope → k8s recommended
  • Atlassian — infra-heavy enterprise origin → expert team assumed
  • GHES — VMware-compatibility origin → weak in non-VMware environments
  • CollabOps — Korean / Japanese public-sector and finance origin → any environment assumed

Tool selection isn't technical comparison. It's a comparison of whose environment each tool assumes.

If you're an IT decision-maker comparing DevOps on-prem tools, the minimum requirements alone across the five tools above show why some don't fit your environment. The most-frequent wrong call — picking a tool that works on 4 vCPU now but forces k8s adoption six months later. That's failing to price the six-months-later migration cost into today's decision.

Tags#onprem#salesforce#gitlab#atlassian#comparison#devops