The Field Mistake Nobody Wants to Admit
You walk into a site visit. The customer says they want 40 IP cameras across three buildings. The facilities manager points at the existing network closet and says "just add them to the main VLAN." IT hasn't been in the conversation yet. The integrator has already priced Verkada and a local NVR option, but nobody has actually calculated how much data moves per day, how long it stays on disk or in the cloud, or whether the uplink can handle failover.
Then go-live happens. The system runs fine for two weeks. On day 15, someone notices that older footage is missing. The NVR ran out of space. Or the cloud system throttles bandwidth and starts dropping frames at 2 PM when the office is busiest. The customer blames the equipment. You're in a margin squeeze trying to figure out what went wrong.
This happens because bandwidth and storage calculations feel like homework. They're not sexy. But they're the difference between a system that works and one that becomes a support ticket every Monday.
Start With Per-Camera Bitrate
Every IP camera publishes a bitrate spec. But the spec is a range, and the actual number depends on resolution, frame rate, and compression settings.
Real-world bitrates for common setups:
- 1080p, 30 fps, H.264 (standard): 3 to 6 Mbps
- 1080p, 30 fps, H.265 (modern codec): 1.5 to 3 Mbps
- 4MP, 30 fps, H.264: 8 to 12 Mbps
- 4MP, 30 fps, H.265: 4 to 6 Mbps
- 8MP (4K), 30 fps, H.264: 15 to 25 Mbps
- 8MP (4K), 30 fps, H.265: 8 to 12 Mbps
These numbers assume motion-adaptive encoding is on. If you turn that off, double them.
Most cameras also have a "keyframe interval" setting (usually 1 to 5 seconds). Shorter intervals mean faster seek times in playback but higher bitrate. Longer intervals save bandwidth but make scrubbing slower. Default is usually 2 seconds. Leave it there unless you have a reason to change it.
The integrator's rule: Ask the customer what they're actually recording for. If it's parking lot theft investigation, 1080p H.265 at 15 fps is plenty. If it's counting people in a retail space or reading license plates, you need 4MP or 8MP at 30 fps. That's not a spec sheet question; it's a requirements question.
Aggregate Bandwidth to the Network
Now multiply per-camera bitrate by the number of cameras and add overhead.
Example: 40 cameras at 1080p H.265 30 fps, averaging 2.5 Mbps each.
40 × 2.5 Mbps = 100 Mbps aggregate stream bitrate.
Add 15% for protocol overhead (RTCP, packet loss retransmission, NTP sync). You're at 115 Mbps.
If you're using a cloud VSaaS platform (Verkada, Eagle Eye, Rhombus), that 115 Mbps leaves the site. Your internet uplink must handle it continuously during business hours, plus some headroom for redundancy and other traffic.
Uplink reality check:
- 100 Mbps sustained needs a 300 Mbps dedicated connection. Most standard business fiber is 500 Mbps / 50 Mbps. That 50 Mbps upload is already spoken for if you're doing cloud video, email, and backups.
- Multi-site customers with 5 locations at 40 cameras each are pushing 575 Mbps egress. That's a dedicated MPLS circuit or multiple carriers.
- If the uplink fails, cloud video stops. On-premise NVR keeps recording. This matters for compliance and incident investigation.
If the site only has cable internet (gigabit down, 35 Mbps up), cloud video at scale doesn't work. You're forced into hybrid (some local NVR, some cloud) or pure on-premise NVR.
Storage Calculation for Retention
Retention is usually the driving cost in surveillance. Insurance, compliance, or incident investigation determines how many days you need to keep footage.
Retention formula:
Daily storage = (Bitrate in Mbps × 86,400 seconds per day) / 8 bits per byte / 1,000,000 bytes per MB
Simplified: Daily storage (GB) = Bitrate (Mbps) × 10.8
Example: 115 Mbps aggregate = 115 × 10.8 = 1,242 GB per day (about 1.2 TB/day).
For 30 days retention: 1.2 TB × 30 = 36 TB raw storage needed.
For 90 days: 108 TB.
But add RAID overhead. RAID 6 (two-disk fault tolerance, standard for NVR) needs 33% extra capacity. 36 TB becomes 48 TB usable.
And add buffer. You need 10-15% free space on a drive for performance. So 48 TB becomes 55-58 TB.
A four-bay NVR with 12TB drives gives you 36 TB raw, maybe 24 TB usable. Not enough for 30 days at this bitrate. You need an 8-bay or a second NVR.
Cloud retention math is different. Verkada, Eagle Eye, and similar platforms charge per camera per month, and retention is baked in (usually 30 days standard, 90 days available). You don't manage storage yourself. But you pay every month, and the total cost over three years is often 2-3x the NVR hardware cost.
Hybrid is a middle path: 7-14 days on local NVR, 30-90 days in the cloud. Reduces bandwidth spend, keeps hot footage local, archives to cloud for compliance.
Network Design Decisions
VLAN Segregation
Video traffic should not share the default office VLAN. Create a dedicated video VLAN (e.g., VLAN 30) for all cameras and NVR/VSaaS gateways.
Why? Broadcast storms from misconfigured cameras can tank your whole network. Video multicast can flood switches. A rogue camera or PoE injector is easier to isolate if it's on its own VLAN.
Segregation also makes firewall rules and QoS easier to manage. See the related post on VLAN segmentation for surveillance and access control for specifics.
PoE Budget
Every PoE port on a switch has a power limit. Most switches support 30W per port (802.3at) or 60W (802.3bt high-power). Modern IP cameras (especially 8MP or motorized domes) draw 15-25W.
40 cameras at 20W average = 800W total. Your PoE switch must support it across all ports, not just per-port. A 48-port gigabit switch with 240W total PoE can't run 40 cameras. You need a dedicated PoE injector or a switch with 600W+ budget.
Always ask: Is the switch budget per-port or per-system? Vendors are vague about this.
Uplink and Redundancy
If you're using cloud video, the internet connection is your single point of failure for live monitoring and new recordings. A local NVR (even small, 2-4 bay) keeps recording if the uplink drops.
For multi-site deployments, each site should have its own local NVR and cloud sync. One site's internet outage doesn't block the others.
Uplink should be at least dual-carrier (fiber + cable, or two fiber providers). Most commercial sites skip this and regret it.
Hybrid vs Pure Cloud vs Pure On-Premise
Pure cloud (VSaaS only):
- No local hardware to manage.
- Automatic updates, AI analytics, easy remote access.
- Depends entirely on internet uplink. If it fails, you lose live view and recording until it's back.
- Monthly recurring cost, no cap.
- Best for small sites (under 16 cameras), good internet, low compliance burden.
Pure on-premise NVR:
- Full control of retention and data.
- Works offline. Recording continues if internet is down.
- Requires PoE switch, storage management, firmware updates on your schedule.
- Higher upfront cost, but no per-camera monthly fee.
- Best for sites with unstable internet, high retention needs (90+ days), or air-gapped networks.
- Day-two support burden on facilities or IT.
Hybrid (local NVR + cloud):
- 7-14 days on NVR, 30-90 days in cloud.
- NVR keeps recording if internet fails. Cloud provides archive and remote access.
- Moderate upfront cost, moderate monthly cost.
- Balances resilience and compliance.
- Most common choice for mid-size commercial sites.
The Integration Checklist
Before you quote or order:
- Get a network diagram. Where is the NVR or gateway? What's the uplink capacity? Are there WAN links between sites?
- Confirm retention requirements. Ask about insurance, compliance (if any), and typical investigation timeline. Don't assume 30 days.
- Calculate aggregate bitrate. Use the per-camera specs from the camera datasheet, not the vendor's marketing number.
- Size the NVR or cloud platform. If on-premise, pick storage and RAID level. If cloud, confirm the uplink can sustain the bitrate.
- Plan the PoE switch. Count total wattage, not just per-port. Confirm redundant power supplies if it's critical infrastructure.
- Segregate the video VLAN. Work with IT to assign a dedicated VLAN and firewall rules.
- Test failover. If the internet drops, does recording continue? Does the customer know how to access local footage?
- Document the configuration. Bitrate, retention days, VLAN ID, NVR IP, cloud platform login. Give it to the customer in writing.
Common Pitfalls
- Underestimating bitrate. Vendors publish specs at lowest quality. Real deployments are 2-3x higher.
- Assuming the uplink is big enough. Fiber gigabit down doesn't mean gigabit up. Check the upload spec.
- Skipping the NVR for small sites. Even a two-camera cloud system benefits from a small local recorder for redundancy.
- Not planning for growth. Spec the NVR and uplink for 50% more cameras than today's count.
- Mixing codecs on one NVR. Some NVRs handle H.264 and H.265 cameras on the same system. Others don't. Confirm before you order cameras.
- Forgetting about heat and power. An 8-bay NVR in a small closet with no cooling will throttle or fail. Plan for ventilation and backup power.
Next Steps
If you're evaluating platforms, start with the network constraints, not the camera features. Bandwidth and uplink will force your hand on cloud vs hybrid vs on-premise faster than any spec sheet. For multi-site deployments, read the network design guide to understand how to scale this across locations. And if you need to segment video traffic from the rest of your network, VLAN design for surveillance walks through the firewall and QoS side.
Get the math right before you order. The rest is implementation.