// INSIGHTS
Streaming From a Venue With an Unreliable Internet Uplink
6 min read
A heritage property outside Jaipur. The banquet manager says there is 100 Mbps. There is one shared broadband line serving reception, the restaurant POS, guest Wi-Fi and a wedding in the adjoining hall. It tests fine at 11am. At 7pm it collapses. This is the normal case, not the bad case. The correct engineering response is to assume the uplink will fail and design so that failure is invisible to the audience.
Design for the uplink you will actually get
The contribution feed, from venue to our MAIN server, is the single most fragile link in the whole chain. Everything downstream sits on redundant infrastructure we control. The venue's internet is not that.
So we build the contribution side around four assumptions:
- The uplink is smaller than advertised. We design a 1080p contribution feed to fit inside roughly 6 Mbps, not 15. Lower contribution bitrate means the transcode ladder at MAIN does more of the work, and the venue link does less of it.
- It will drop at least once. Ingest runs on two independent paths with automatic failover. When one path stops delivering, the other is already carrying the same feed and the cut happens without operator intervention.
- Encoders handle drops badly by default. A 25-second socket idle timeout is what forces a clean reconnect. Without it, an encoder that loses connectivity can sit in a half-open socket, believing it is streaming, sending nothing. That is the failure mode that produces a dead player with no alarm anywhere.
- The recording is not the stream. A local ISO recording runs on-site throughout, independent of connectivity. Whatever happens to the network, the archive is complete.
Contribution-side design targets
The on-site kit
Bonded uplink
A wired line plus modems on two different carriers, bonded into one logical connection. Traffic is spread across all of them continuously rather than switched over after a failure.
Dual-path ingest with automatic failover
Two contribution streams reach MAIN by separate routes. The platform picks the healthy one. No operator has to notice and react.
Error-correcting transport
Contribution runs over a transport with retransmission and configurable latency buffer, so brief packet loss is repaired rather than shown to viewers as artefacts.
Local ISO recording
Every camera and the programme mix record locally to disk on-site. The archive never depends on the network, which also means a total outage still leaves you with an on-demand asset.
Site work before the event
Test at the hour the event runs
Speed-test the venue line at 7pm on a weekday, not at 11am on a Tuesday. Also test from the actual position the encoder will sit in, not from the business centre.
Order a dedicated line three weeks out
Provisioning a temporary fibre or leased line in India realistically takes two to four weeks, and longer at heritage or rural properties. This is a lead-time problem, not a budget problem.
Check mobile signal in the room, not the lobby
Basement ballrooms and thick-walled heritage halls kill cellular. If signal is poor, plan a cable run to a roof or courtyard position and get written venue permission days ahead.
Rehearse the failure
During rehearsal, physically unplug the primary line while streaming. Watch what the player does. If nobody in the control room notices within ten seconds, the monitoring is wrong.
What usually goes wrong
The venue's stated bandwidth is usually a download figure. A property advertising 100 Mbps commonly has 100 down and 10 up, and upload is the only number that matters for streaming. It is also almost always contended, meaning your 10 Mbps is shared with guest Wi-Fi and a POS system that decides to sync at exactly the wrong moment. We do not argue about this any more. We test, write down the real measured upload figure, and design the contribution bitrate to fit inside about 60% of it. The remaining headroom absorbs the wedding next door.
The less obvious failure is correlated redundancy. Two SIMs from two carriers look like independence until you discover both towers serving that hall are on the same congested backhaul, or that the basement kills both equally. Bonded cellular does not help in a place with no cellular. When a site survey shows this, the honest answer is that the fix is physical: a cable run to a rooftop antenna position, or a temporary line ordered weeks in advance. Both need venue cooperation and time. Neither can be solved by turning up on the day with better equipment, and we would rather tell you that at the survey than at 6pm on show day.
Common questions
What do viewers see if everything drops at once?
A branded holding card with a short message, rather than a dead or spinning player. The player stays connected to the edge and picks the feed back up automatically when contribution returns. Meanwhile the local recording keeps running, so the segment that viewers missed still exists for the on-demand version.
Can you stream on mobile connectivity alone?
Yes, with two carriers bonded and a realistic budget of around 8 to 10 Mbps of usable upload in a location with good signal. It is a normal configuration for outdoor and rural events. The condition is a site survey that confirms signal strength at the encoder position at the time of day the event runs.
Who pays for a temporary internet line?
Usually the client, as a pass-through cost in INR, because the connection is ordered in the venue's or organiser's name. We specify what is needed, coordinate with the provider and test on delivery. Budget for it early, since the constraint is provisioning time far more than price.
Streaming from a difficult venue
Send us the venue name and date and we will run a connectivity survey before anyone commits to a stream.
