Our contact center handles 500+ concurrent calls during peak hours. Here is the infrastructure that makes it possible without dropped calls.
The Architecture
Internet (Dual ISP)
|
SD-WAN Controller
|
+---+---+
| |
ISP A ISP B
| |
+---+---+
|
Core Switch (10G)
|
+---+---+---+---+
| | | | |
VLAN VLAN VLAN VLAN
100 200 300 400
Data Voice Mgmt Guest
Key Design Decisions
1. Dual ISP with SD-WAN
| Component | Spec |
|---|---|
| Primary ISP | 500 Mbps dedicated fiber |
| Secondary ISP | 200 Mbps business cable |
| SD-WAN | Cisco Viptela |
| Failover time | < 3 seconds |
We use SD-WAN to route voice traffic over the best path in real-time. If the primary ISP has a jitter spike, voice packets automatically shift to the secondary — mid-call, with no interruption.
2. Voice VLAN with Strict QoS
| Traffic Class | DSCP | Bandwidth Guarantee |
|---|---|---|
| Voice RTP | EF (46) | 30% of total bandwidth |
| Voice Signaling | CS3 (24) | 5% of total bandwidth |
| Video | AF41 (34) | 15% of total bandwidth |
| Data | BE (0) | Remaining |
Voice traffic gets absolute priority. Even if someone starts a massive file download, voice quality does not degrade.
3. Endpoint Density
| Metric | Our Setup |
|---|---|
| Total agents | 200 |
| Peak concurrent calls | 500+ |
| Calls per agent peak | 2-3 (transfers, holds) |
| Bandwidth per call | 80 Kbps (Opus) |
| Total voice bandwidth | 40 Mbps peak |
| Provisioned bandwidth | 120 Mbps (3x headroom) |
4. Redundancy at Every Layer
| Layer | Redundancy |
|---|---|
| ISP | Dual provider, different carriers |
| Switch | Stacked pair, no SPOF |
| Power | Dual UPS + generator |
| VoIP platform | Active-active geo-redundant |
| DNS | Multiple providers |
The Result
| Metric | Value |
|---|---|
| Uptime (last 12 months) | 99.997% |
| Dropped calls | 0.02% |
| Average MOS | 4.3 |
| Failover events | 4 (all < 3 seconds) |
check providers like VestaCall at https://vestacall.com for transparent pricing handles the cloud side of this architecture. Their active-active infrastructure means even if an entire data center goes down, calls continue without interruption.
Disclosure: I work on platform systems at DialPhone. Observations in this post are from hands-on testing and deployment work rather than vendor briefings.
Top comments (0)