Measuring ISP Link Quality Across India: Data from 900+ Sites

Mar 26, 2026

Hopbox team

When you operate 900+ SD-WAN devices across India, each with multiple WAN uplinks, you end up sitting on a dataset that most people never get to see: continuous, granular, real-world ISP link quality measurements from hundreds of cities and towns, 24/7.

This post shares what that data looks like — the methodology behind our measurements, regional patterns we’ve observed, ISP-level trends, and what it all means for anyone designing multi-WAN network architectures in India.

Methodology

What We Measure

Every Hopbox CPE device runs a lightweight probe daemon that performs the following measurements against our regional hub servers every 30 seconds:

Metric	Method	Target
Latency (RTT)	ICMP echo	Regional hub
Jitter	UDP stream (50 packets, 20ms interval)	Regional hub
Packet loss	ICMP echo (100 packets)	Regional hub
DNS resolution time	Recursive query	Public DNS + local recursor
HTTP reachability	GET request	Known endpoint

All measurements are per-WAN-link. A site with 3 WAN links generates 3 independent sets of measurements.

Data Collection and Aggregation

Measurements are collected every 30 seconds and exposed as Prometheus metrics.
Regional Prometheus instances scrape and store raw data with 15-day retention.
5-minute aggregations (mean, p50, p95, p99) are computed via recording rules and federated to our global Prometheus instance with 90-day retention.
For this analysis, we use 5-minute aggregated data over a 6-month window.

Scale of the Dataset

Sites: 900+
WAN links monitored: ~2,400 (average 2.6 links per site)
Data points per day: ~140 million (across all metrics and links)
Observation period: 6 months

Regional Patterns

India’s network infrastructure varies dramatically by geography and urbanization level. We categorize our sites into three tiers:

Metro Cities (Mumbai, Delhi, Bangalore, Chennai, Hyderabad, Kolkata)

Metric	Fiber (p50)	Broadband (p50)	4G Backup (p50)
Latency	3-8 ms	8-20 ms	25-60 ms
Jitter	< 1 ms	1-3 ms	5-15 ms
Packet loss	< 0.1%	0.1-0.5%	0.5-2%
Availability	99.8%+	99.2-99.6%	98-99%

Metro fiber is predictably excellent. Enterprise-grade fiber from major providers consistently delivers sub-10ms latency to our regional hubs with negligible packet loss. The story changes significantly with broadband connections, where shared last-mile infrastructure introduces variability.

Tier-2 Cities (Pune, Ahmedabad, Jaipur, Lucknow, Kochi, Chandigarh, etc.)

Metric	Fiber (p50)	Broadband (p50)	4G Backup (p50)
Latency	8-15 ms	15-40 ms	30-80 ms
Jitter	1-2 ms	2-8 ms	8-25 ms
Packet loss	< 0.2%	0.2-1%	1-3%
Availability	99.5%+	98.5-99.2%	97-99%

The gap between fiber and broadband widens noticeably in Tier-2 cities. Fiber availability is still good but not metro-level. Broadband connections show significantly more variance, particularly during peak hours.

Tier-3 Cities and Towns

Metric	Fiber (p50)	Broadband (p50)	4G Backup (p50)
Latency	12-25 ms	25-80 ms	40-120 ms
Jitter	2-5 ms	5-20 ms	15-40 ms
Packet loss	0.2-0.5%	0.5-3%	2-5%
Availability	99-99.5%	96-98.5%	95-98%

This is where multi-WAN becomes essential, not optional. No single link type in Tier-3 locations delivers the reliability needed for business-critical applications. The combination of two mediocre links, however, can achieve effective availability above 99.9%.

ISP Comparison (Anonymized)

We won’t name ISPs directly, but we can share generalized observations based on provider categories:

Large National Telcos (Fiber)

Most consistent performance across regions
Latency variance typically under 2x between metro and Tier-3
Outages tend to be short (< 30 min) but can affect large areas simultaneously
Best SLA adherence

Regional Broadband Providers

Highly variable between providers and even between cities for the same provider
Often the best price-to-performance in their home region
Outages can be prolonged (hours) due to smaller NOC teams
Peak-hour congestion is a common pattern (18:00-23:00 IST)

4G/LTE Providers (Backup Links)

Latency is acceptable for failover but not ideal for primary use
Significant congestion patterns in dense urban areas
Rural coverage gaps still exist
Data caps make them unsuitable as primary links for most deployments

Time-of-Day Patterns

One of the most consistent findings across our dataset: broadband link quality degrades predictably during evening hours.

Broadband Latency by Hour (all sites, p50)

00:00 ████████░░░░░░░░░░░░  12 ms
04:00 ██████░░░░░░░░░░░░░░  10 ms
08:00 ████████░░░░░░░░░░░░  12 ms
12:00 ██████████░░░░░░░░░░  15 ms
16:00 ████████████░░░░░░░░  18 ms
18:00 ██████████████████░░  28 ms
20:00 ████████████████████  32 ms
22:00 ████████████████░░░░  24 ms

This pattern is driven by residential broadband usage in the same last-mile infrastructure. Fiber connections show much less time-of-day variation.

For Hopbox sites, this means our SD-WAN link-quality monitoring and automatic failover is most active between 18:00 and 23:00 IST, shifting traffic from degraded broadband links to healthier alternatives.

Monsoon Season Impact

We track a noticeable degradation in link quality during the monsoon season (June-September), particularly for:

Broadband connections in areas with aerial cabling: packet loss increases 2-5x during heavy rain events.
4G backup links: signal degradation during intense rainfall causes intermittent connectivity.
Fiber connections: largely unaffected (underground), but fiber cuts due to construction/flooding do occur.

The practical impact: during monsoon months, sites in affected regions experience 30-50% more failover events. Our multi-WAN architecture handles this gracefully — most failovers are invisible to end users — but it underscores why single-link deployments are risky in India.

Fiber vs Broadband vs 4G: Cost-Performance Analysis

The right link mix depends on the site’s tier, application requirements, and budget:

Link Type	Typical Cost (INR/month)	Best For	Limitations
Enterprise Fiber	8,000 - 25,000	Primary link, VoIP, video	Availability in smaller towns
Business Broadband	1,500 - 5,000	Secondary link, bulk data	Peak-hour congestion
4G LTE (SIM)	500 - 2,000	Failover backup	Latency, data caps

Our standard deployment recommendation:

Metro sites: 1 fiber (primary) + 1 broadband (secondary)
Tier-2 sites: 1 fiber (primary) + 1 broadband (secondary) + 1 4G (backup)
Tier-3 sites: 2 broadband from different ISPs (primary/secondary) + 1 4G (backup)

The SD-WAN overlay continuously measures all links and routes traffic to the best-performing path, making the effective quality better than any individual link.

Implications for Multi-WAN Strategy

The data consistently supports a few principles:

No single ISP is reliable enough everywhere. Even the best national providers have regional weak spots. Multi-WAN with ISP diversity is not a luxury.
Fiber availability is the single biggest predictor of site quality. Sites with at least one fiber link have dramatically better median performance than broadband-only sites.
4G is a viable backup, not a viable primary. It fills the gap during outages but can’t sustain the consistency needed for business applications.
Active link monitoring changes the equation. With real-time quality measurement and automatic failover, two mediocre links can deliver better effective quality than one good link. This is the core value proposition of SD-WAN.
Plan for monsoon. If you’re designing network infrastructure for India, build your availability targets around monsoon-season performance, not dry-season performance.

Data Access

We are evaluating how to make a subset of this data publicly available — an “ISP quality index” for India based on real SD-WAN probe data. If that’s something you’d find useful, let us know.

This analysis reflects 6 months of continuous measurement across 900+ sites. Network quality in India has improved significantly over the past few years, particularly with the expansion of fiber infrastructure. But the variance between locations, providers, and conditions remains large enough that multi-WAN architectures with intelligent failover aren’t optional for anyone running distributed operations across the country.

v1.7.9