NetPulse vs SiteWatch — Alarm Correlation Analysis

Date: 2026-02-18 • Alarm: Technology Wise Site Down • Full 24-Hour Analysis

NetPulse Alarm Records
Event-based (1 row per alarm)
SiteWatch Total Rows
Polling-based (1 row per 15-min batch)
SiteWatch Unique Events
Deduplicated by source_event_time
Common Sites
NetPulse-Only Sites
96.9% had alarms <15 min
SiteWatch-Only Sites
All had alarms from before Feb 18

Site Overlap

Venn-style breakdown of sites appearing in each system

Alarm Count Comparison (Common Sites)

How many sites match vs differ between systems

Row Count Comparison: Raw vs Meaningful

SiteWatch has 3.4x more rows but only 0.53x the unique alarm events

79.4%
NetPulse Alarms < 15 min
7,977 of 10,045 alarms
15 min
SiteWatch Polling Interval
95 batches across the day
5.4x
Avg Rows Ratio
SW rows per NetPulse alarm (inflation)
0.82
Unique Events Ratio
SW unique events per NetPulse alarm

NetPulse Alarm Duration Distribution

Most alarms clear within 15 minutes — invisible to SiteWatch polling

Difference Distribution (NetPulse − SiteWatch)

Per-site difference in alarm count — positive = NetPulse sees more

NetPulse Alarms Per Site Distribution

Most sites have 1-3 alarms; flapping sites can have 100+

Primary Cause: Polling Granularity Mismatch

SiteWatch polls every 15 minutes. NetPulse records every raise/clear cycle. Since 79.4% of NetPulse alarms last less than 15 minutes, they fall entirely between two SiteWatch polling windows and are never captured.

Secondary Cause: Alarm Flapping

NetPulse records every re-occurrence separately. Site AGQ8168 has 176 NetPulse records but only 4 SiteWatch events. 951 sites show duplicate FirstOccurred timestamps, totaling 4,401 flapping records.

SiteWatch Row Inflation

SiteWatch repeats each active alarm across polling batches. An alarm active for 4 hours generates ~16 rows. This inflates SiteWatch's raw count to 33,955 rows for only 5,298 unique events.

Legacy Alarm Carryover

63 SiteWatch-only sites have alarms originating before Feb 18 (some from Jan 6). NetPulse scopes to alarms with FirstOccurred on Feb 18, so these long-running alarms are excluded from NetPulse.

Hourly Alarm Distribution — NetPulse vs SiteWatch

NetPulse = new alarms raised per hour • SiteWatch = unique active sites per hour

16:00 Spike (NetPulse)

A massive spike of 1,860 alarms at 16:00 in NetPulse — over 3x the hourly average. SiteWatch shows a smaller peak (~985 sites), confirming most of these were very short-lived flaps.

01:00 Overnight Spike

Both systems show a secondary spike at 01:00 (~865 alarms). This suggests a scheduled nightly process (maintenance window, batch job, or power cycling) affecting many sites.

Afternoon Buildup

SiteWatch shows a gradual rise from 13:00–16:00, reflecting alarms that accumulate and persist. NetPulse shows sharper spikes, capturing the rapid fire-and-clear pattern typical of transient issues.

Region-wise Match Rate

Percentage of common sites where alarm counts match, NetPulse is higher, or SiteWatch is higher

Region Summary Table

Click column headers to sort

RegionTotal SitesExact MatchMatch % NetPulse HigherNetPulse Higher %SW HigherSW Higher %

South Region — Worst Alignment

Only 56% match rate with 41% showing NetPulse higher. South has the most sites (1,773) and the most alarm flapping, suggesting infrastructure instability in southern regions.

Central A — Best Alignment

79% match rate — best among all regions. Fewer transient alarms indicate more stable infrastructure or less sensitive alarm thresholds.

Pearson Correlation
NetPulse alarm count vs SW unique events
63.0%
Exact Match Rate
2,051 of 3,254 common sites
34.6%
NetPulse Sees More
1,126 sites (flapping/short alarms)
2.4%
SiteWatch Sees More
77 sites (legacy carry-over)

Scatter: NetPulse Alarm Count vs SiteWatch Unique Events (per site)

Points on the diagonal line = perfect agreement • Points above = SiteWatch higher • Points below = NetPulse higher

Full Site Comparison Table

Search, filter, and sort all sites across both systems

Site ID Region City NetPulse Alarms SW Events SW Rows Difference Short (<15m) Status

1. The Count Difference is Caused by Polling Granularity

79.4% of NetPulse alarms last < 15 minutes

SiteWatch's 15-minute polling window misses roughly 40% of alarm events that NetPulse captures. Any alarm that raises and clears within a single polling cycle is invisible to SiteWatch. This is the single biggest driver of the count disparity.

2. SiteWatch Inflates Row Counts but Deflates Event Counts

33,955 rows → 5,298 unique events

SiteWatch records the state of each active alarm at every 15-minute batch (95 batches/day). A single alarm active for 6 hours generates ~24 rows. The raw row count is 3.4x higher than NetPulse, but unique events are only 0.53x.

3. Neither System is "Wrong"

Different questions, different answers

NetPulse answers: "How many times did alarms fire?" (event-centric, captures every flap)
SiteWatch answers: "How many alarms were active at each time window?" (state-centric, captures duration)

4. For True Correlation: Use source_event_time

63% exact match when using unique events

Compare NetPulse alarm counts against SiteWatch's unique source_event_time per site, not raw row counts. This gives 63% exact match for common sites, and the remaining differences are primarily short-lived flapping alarms.

5. Flapping Sites Need Investigation

951 sites show alarm flapping patterns

Sites like AGQ8168 (176 NetPulse records vs 4 SW events) indicate unstable network elements that go down and come back within minutes, repeatedly. These are candidates for proactive maintenance. The South region has the highest concentration of flapping sites.

6. SiteWatch Captures Long-Running Alarms NetPulse Misses

63 sites with alarms originating before Feb 18

SiteWatch snapshots all currently-active alarms regardless of origin date. NetPulse scopes to FirstOccurred on the analysis date. For a complete picture, combine both views: NetPulse for new events, SiteWatch for ongoing state.