Skype for Business 2015 is the on-premises UC platform our larger clients still standardize on in 2016. Multi-tenant and 10,000-user designs get the glory later; most enterprises need a solid single-organization topology first. This post is the design decisions that age well: front-end pools, edge, HA, capacity, and what we refuse to undersize.
Topology simplicity
Prefer a standard edition only when the business accepts the tradeoffs. Enterprise pools with proper SQL backends are the default for anything that must survive patch day. Pair front ends, plan for quorum, and document failover. “We’ll add HA later” is how you run single points of failure for three years.
Edge and external access
Edge services, reverse proxies, certificates, and DNS are half the project. Split-brain DNS, missing SRV records, and expired certs cause more “Skype is down” tickets than media codec debates. We treat edge as a product with owners and monitoring, not a weekend install.
Capacity and concurrency
Size for peak concurrent, not licensed seats. Call centers and classroom training days lie to average-based models. Load test where politics allow; otherwise use conservative concurrency ratios and leave headroom.
HA beyond the front-end pool
File stores, SQL, office web apps if used, mediation/SBA for voice — each has a story. Voice especially: PSTN gateways and SBCs fail independently of the SfB pool. Map dependencies before you promise 99.9%.
Client and channel discipline
Client versions, GPO for updates, and VDI media settings matter for perceived quality. A perfect pool with broken clients is still a failed UC project.
Monitoring
QoE databases, call quality methodology, and SCOM packs — without them you fly blind. “Users say audio is bad” is not an RCA.
Voice as a separate program
Presence and chat can go live while voice is still in design. Pretending they are one weekend is how you ship broken PSTN. Mediation servers, gateways, dial plans, emergency calling requirements, and SBC vendors each need owners. If voice is in scope, it gets its own project plan and test matrix — not a bullet under “UC phase 2.”
Persistent chat and compliance
Some industries require retention and eDiscovery patterns that outgrow default assumptions. Involve legal early. Persistent chat architecture choices are expensive to reverse after adoption.
Multi-site and WAN
Centralized pools with remote sites depend on WAN quality for media. Branch office appliances and bandwidth policies are design inputs. We map site counts and typical call volumes before freezing topology.
Upgrade and cumulative update discipline
SfB farms that skip CUs become unsupportable. Patch cadence, test users, and rollback notes belong in the runbook from day one. “Never touch production” is not an HA strategy; it is technical debt with a calendar.
Relationship to VDI
Published SfB clients inside XenDesktop or RDS need media optimization strategies. Without them, every VDI user becomes a bandwidth and CPU complaint. Coordinate EUC and UC designs; siloed ownership produces “both teams say the other stack is fine.”
Scenario walkthrough
Consider a mid-size organization with hybrid identity, mixed desktop delivery, and a mandate to reduce risk without stopping the business. Week one is inventory and sponsor alignment. Week two is a written target state with two options and explicit out-of-scope items. Weeks three and four are pilot build and measurement. Only then does broad change begin. Compressing that sequence into a single weekend is how outages are born.
Along the way, three conversations dominate: who owns identity decisions, who pays for platform capacity, and what residual risk leadership accepts in writing. When those conversations are avoided, engineers improvise under pressure and the organization inherits accidental architecture.
We keep a living risk register with severity, mitigation, residual risk, and owner. The register is reviewed in the same meeting as the delivery burn-down. Risks that never move owners are the ones that become incidents.
If you're facing this
If you are standing up Skype for Business 2015 as a multi-year platform, design HA and edge first. We architect SfB estates that hold under real concurrency — bring peak call estimates and whether voice is in scope on day one.