Home › SysPeep Monitor › Why SysPeep › Honest Numbers
03 · Honest numbers

The monitor that never invents good news


An uptime figure is a claim, and somebody will eventually ask you to defend it.


Key points

  • Every availability figure is bounded by the coverage that was actually measured.
  • A month with no monitoring data shows as no data, never as 100% uptime.
  • A single missed poll is a red gap on the graph, and every gap says why: the device did not answer, collection overran, or SysPeep itself was not observing.
  • Availability is computed over the time SysPeep was actually running and polling; a stopped monitor is a hole in the record, not uptime and not downtime.
  • Countdowns — certificate days, disk space, the monitor's own storage — are shown as projected dates.
  • The monitoring server and every probe are monitored devices in their own right.

On this page: Bounded by coverage Not observing Graphs that show the gap Degraded is not down Countdowns become dates Reports The monitor watches itself


The awkward truth about a great deal of monitoring software is that it computes availability over the window you asked for, whether or not it was actually watching for all of it. A month with no monitoring data comes out as 100%. A run of missed polls is drawn as a smooth, reassuring line. A collector that quietly stopped on the 3rd produces a flawless report on the 31st.


SysPeep was built on a different commitment: it never shows a number the data did not earn. That sounds like a small thing until you are standing in front of the CFO, or the ISP, or the application team that insists it was the network.


Availability bounded by what was measured


Every availability figure in SysPeep is limited to the coverage that was actually recorded. A run history records when SysPeep was running and polling, and availability, downtime, outage counts and period comparisons use that covered time only. Uncovered periods render as no data, never as uptime. Partial months are labelled with how much of the month was measured, so "99.8% of a month we watched 60% of" cannot be mistaken for "99.8%", and a period-over-period arrow gives way to "not comparable" when either period is under 99.5% measured. If the monitor was blind, the report says so, in the place where the number would have been.


A month SysPeep did not watch shows as no data — never as 100% uptime.

When it is not watching, it says so


When the service is stopped, starting up or network-blind, the dashboard's tiles turn grey and hatched with "Not observing since", and the tree and the map say their statuses are frozen. A reading older than three intervals is greyed and dated "as of" the time it was taken, and the dashboard says how much of the last 24 hours was observed. Nobody mistakes the last known state for the current one.

Graphs that show the gap


A missed poll is drawn as a red no-response band on the graph — even a single missed interval — and the band says why the poll is missing: the device did not answer, or collection took longer than the interval, with the cause in its title. Time when a sensor waited for a slower one, or when SysPeep itself was not observing, is drawn as a separate neutral hatch rather than as the device's fault. An outage never looks like a smooth line, and a flapping link looks like exactly what it is. The graph is allowed to be ugly, because an ugly graph that tells the truth is worth more than a pretty one that does not. (The pretty one is the one that gets you blamed later.)


Every reading is one click from the chart: a Readings table lists the whole range, newest first, in pages of a hundred, and a click on a chart point opens its row. The exact value behind a spike is a click away, and so is proof of what was recorded.


Degraded is not down


Each device has a chosen sensor that proves reachability, and only that sensor can mark the device down. A full disk, a stopped service or a slow database makes the device degraded, which shows on the dashboard, raises its own alert, and leaves the availability figure alone. Down keeps meaning unreachable, so the availability number on the report means what the reader assumes it means.


Countdowns become dates


"Forty-one days" is a number. "Expires on the 14th of next month" is a date somebody can put in a calendar. SysPeep turns countdown metrics — certificate days remaining, disk days until full, the monitor's own storage — into a projected date, shown beside the current value and the worst value seen. Capacity exhaustion is forecast rather than discovered.


The question in the meetingWhat SysPeep shows
"What was our availability last month?"The figure, bounded by measured coverage, with partial coverage labelled as such
"Was that outage really twenty minutes?"A red no-response band on the graph with the exact interval, even for a single missed poll
"Why does the report say the server was up when the app was down?"The device was reachable (up) and degraded (service failed); both are on the record, separately
"When does the wildcard certificate expire?"A date, next to the current days-remaining value and the worst value seen
"When will the monitoring server run out of disk?"A projected date from the built-in SysPeep server device
"Was the monitor even running on Sunday?"The run history: the dashboard says how much of the last 24 hours was observed, and an unobserved Sunday is a hole in the record, not uptime
"What exactly was the value at 02:14?"The Readings table, opened from that point on the chart
"What happened right before it broke?"The events timeline: every syslog line, trap, Windows event, availability change, probe event and flow notice, with what each became

Reports you can hand upward


Every report inherits the coverage rule above, which is what makes it safe to forward: executives get numbers you can stand behind, and the ISP gets evidence rather than an opinion. In the paid editions reports can be scheduled and delivered by email, so the monthly report arrives without anyone building it in a spreadsheet on the last Friday of the month.


The dashboard carries fleet KPIs, a site latency and loss heatmap, hygiene nudges, widgets such as top talkers and traffic by exporter, and a NOC wall view that works on any spare screen. The morning check takes about thirty seconds, which is the right length for a morning check.

The report set

  • Executive scorecard
  • Health, SLA and alerts
  • Capacity, with projected dates
  • Inventory and traffic
  • CSV export; PDF export (paid editions)
  • Scheduled email delivery (paid editions)

The monitor watches itself


Monitoring systems are famous for being the one thing nobody monitors. SysPeep ships with a built-in device for the SysPeep server itself, in its Self-Monitor group, and every remote probe is likewise a monitored device with its own silence, status, CPU, memory and spool sensors. A dead probe is an incident, not a mystery of missing data that someone notices three weeks later from a suspiciously flat graph.

  • Collection failures and timeouts, and the poll cycle's duration, spacing and late starts
  • Event rate, host CPU and memory, and database latency
  • Disk usage on the data volumes, with a days-until-full date
  • Retention you set: metric history and events share one horizon (90 days by default), flow history has its own, and a storage ceiling bounds the flow data

You can add sensors to any product. You cannot add honesty afterwards, because by the time you discover the report was computed over a month the collector was down, the report has already been sent. The design decision has to be made before the first poll, and in SysPeep it was.


Run the trial through a month you will have to report on

Install it, let it watch for a few weeks, and compare its availability report with the one you currently send. Any difference between the two is worth a hard look.

Download Free TrialNext: Everything it monitors