Solution · Uptime and availability

Uptime monitoring that can say "I was not watching"


The uptime report said 100% for a month the collector was not running. Everyone in the room knew it, and nobody could prove otherwise. SysPeep was built so that number cannot be printed.


Key points

  • ICMP, TCP port, DNS query, NTP clock and SIP OPTIONS checks for devices; HTTP sensors and multi-step web journeys for services, with certificate days remaining shown as a date.
  • You choose which sensor proves a device is reachable. Only that sensor can mark it down; a full disk or a stopped service marks it degraded and leaves the availability record alone.
  • Availability is computed over observed time only. A month with no data shows as no data, never as 100%; a missed poll is a red gap on the graph.
  • In the paid editions a sensor can be checked from up to four places, this server and remote probes, and every alert names the place that raised it.
  • The monitor keeps a run history of itself, so a stopped monitor is a hole in the record rather than uptime.

On this page: The checks Down versus degraded Honest availability From more than one place SLA reports What it is not


Uptime is the number everyone outside the IT department understands, which is exactly why it is the number most often misreported. Any tool can ping a host and count the answers. The hard parts are deciding what "down" means, admitting when the tool itself was not looking, and producing a figure that survives a sceptical reader. Those are the parts SysPeep's design decisions are about.


The checks


CheckWhat it proves
ICMP, TCP port, DNS query, NTP clock, SIP OPTIONSReachability, service ports, resolver health, clock drift and VoIP signalling: the "is it up, and is it answering correctly" checks that catch the cheap failures before users report them
HTTP sensors: status, latency, JSON and text matching, certificate days remaining, minimum TLS version, weak ciphers, security headersA web service checked the way an auditor would check it, with the certificate expiry shown as a date instead of a countdown someone has to convert
Multi-step web journeys: log in, navigate, verify, scripted end to end"The site is up but nobody can log in" is caught as an outage, because it is one
Mail flow round trips, LDAP bind, RADIUS authenticationThe failures that lock the whole company out at 8:55 a.m., caught by the monitor instead of the helpdesk queue
Windows services over WMI, Linux over SSH, hardware over IPMI and RedfishThe server is up but the service is not; the operating system is down but the fan failure was seen first over the management port

Down means unreachable. Everything else is degraded.


You choose which sensor proves a device is reachable, or let SysPeep choose. Only that sensor can mark the device down. A full disk, a stopped service, a slow query or an expiring certificate marks it degraded, with its own alert, and leaves the availability record alone. The uptime figure therefore measures what the word means, and the degraded state carries the problems that deserve a ticket rather than a page.


A month SysPeep did not watch shows as no data, never as 100%. Partial months are labelled with how much was measured.

Availability over observed time only


Every availability figure is bounded by the coverage that was actually measured. A missed poll is drawn as a red gap on the graph with the reason beside it, not interpolated into a flat line. Devices behind a failed upstream are marked unreachable under the upstream's incident, so a dead switch does not become forty outages in forty availability records. If the monitoring host loses its own network link, judgement freezes, exactly one incident is raised, and no device is falsely marked down.


The monitor also keeps a run history of itself. A built-in device for the SysPeep server, in the Self-Monitor group, records when the service was running and polling, so a stopped monitor is a hole in the availability record rather than a month of silent uptime. The honest numbers page goes through every rule.

What the figure will not do

  • Report 100% for time it did not watch
  • Count a full disk as downtime
  • Count forty hosts behind a dead switch as forty outages
  • Hide a missed poll in a flat line
  • Forget that the monitor itself was off

Checked from more than one place


In the paid editions, remote probes poll a branch, a DMZ or a segmented network from the inside and report back over mutual TLS, and a sensor can be checked from up to four places, this server or any probe, with the device still counted once. The console says which place saw and reached each device, and every alert names the place that raised it, so "the web server is down" and "the web server is unreachable from the branch" are different incidents with different causes.


SLA reports you can stand behind


SysPeep produces an executive scorecard and health, SLA, alerts, capacity, inventory and traffic reports, exportable as CSV in every edition and as PDF in the paid editions, where they can also be scheduled and delivered by email. Because the figures are coverage-honest, an SLA report is something you can hand to the person who signs the renewal or disputes the ISP's bill. The library's uptime and availability guides cover what "up" really means and how to measure availability people can trust.


What it is not


SysPeep runs on your own Windows server and checks from your own network and your own probes. It is not a hosted service checking your public website from twenty cities on three continents. If what you need is outside-in monitoring of a public site from around the world, that is a different kind of product. If what you need is to know whether the servers, services and devices your users depend on are actually up, measured from where the users are, and to be able to prove it afterwards, this is the one.


Questions administrators ask about uptime


Does it monitor internet connection uptime and ISP uptime?

An ICMP or HTTP sensor against the ISP's next hop and a public address, polled every 30 seconds, is an internet uptime monitor with the outage record to match; in the paid editions a QoS packet train between two probes measures the circuit itself. The ISP accountability series covers turning that record into a credit.


Is there an uptime SLA calculator? What do 99.9% and 99.99% allow?

Over a year, 99.9% (three nines) allows about 8.8 hours of downtime, 99.99% (four nines) about 53 minutes, and 99.999% about 5 minutes. The SLA report computes the achieved figure over observed time only, so it is the number you can put beside the promise.


Which website monitoring tools does it replace?

The internal ones: checks that your sites, portals and web applications answer correctly, measured from where your users are. It does not replace a hosted service that checks a public site from twenty cities, and the page says so above.


How can uptime be improved?

By finding the causes of the downtime you have, which is what the events timeline beside each outage is for: the switch's syslog line, the failed poll, the certificate that expired. Most uptime is lost to a few repeating causes, and the SLA and alerts reports make them visible.


Does it monitor Apache HTTP server, IIS and web server performance?

HTTP sensors measure status and latency against any web server, multi-step journeys check the pages behind the login, and the host's process and CPU counters come over WMI or SSH. Apache HTTP server monitoring and IIS monitoring are the same sensors on different hosts.


Is latency the same as ping? Can I have low ping but high packet loss?

Ping reports round-trip latency, one number per packet; loss is the share of packets that never came back. They are independent: a link can answer quickly most of the time and still drop one packet in fifty, which is why both are graphed and both have thresholds.


How do I check system uptime with a PowerShell command, get the uptime on Linux with the server uptime command, and why monitor it instead?

On one Windows server, PowerShell's (Get-CimInstance Win32_OperatingSystem).LastBootUpTime gives the last boot; on Linux the uptime command, or cat /proc/uptime, does the same. Those tell you how long the box has been up, not whether the services on it were reachable, and only for the machine you are logged on to. SysPeep polls every device on a 30-second clock and reports coverage-honest uptime: the share of the period the check actually passed, with the minutes it could not observe shown as unknown rather than counted as up.


What makes the best website monitoring tools, and how does SysPeep check websites?

The best website monitoring tools check the page the way a visitor does, from outside the server, and tell you why it failed: HTTP status, response latency, expected content, certificate days remaining and TLS settings, plus the pages behind a login. SysPeep's HTTP sensors do that per URL and its multi-step journeys follow the login; both count against the device the site runs on, not as extra licenses.


Monitor five things that matter, free, and read the first honest report

Download SysPeep, add the firewall, the core switch, the server, the web app and the printer everyone blames, and let it run for a month. Up to 5 devices are free at full depth for as long as you like; the 30-day trial unlocks every feature except clustering.

Download Free TrialReports you can stand behind