Infrastructure monitoring is not one market. It is six, and products routinely cover several of them at different depths, which is why feature comparisons between two tools frequently compare things that are not alternatives.
Before any product name is useful, the question is which class the problem belongs to.
The six
Availability checking asks whether something is reachable from outside. Cheap, easy to run, and the only class that keeps working when everything else is broken. It cannot tell you why, and it measures the path as much as the destination.
Metrics and time series ask how much of something there is and whether it is changing. This is the basis of most alerting and nearly all capacity work. It cannot explain a single request, and its costs are governed by the cardinality article in the infrastructure section.
Log management asks what exactly happened, in order, with the specifics. It answers the questions nobody predicted, and its index is where the bill comes from.
Tracing and code-level tooling asks where one operation spent its time across the services it touched, and in the deeper products, which function inside them. Only this class answers that question once a request crosses a boundary, and it requires context to be propagated everywhere.
Front-end and real user measurement asks what the person on the other end actually received, including everything that happened after the server was finished. The article on synthetic and real user data covers what it can and cannot see.
Network and device monitoring asks which link, port, appliance or route is at fault, usually by polling management interfaces on hardware that cannot run anything. It is a different discipline with different vocabulary and different products.
Naming the decision you cannot currently make
The same test as in the workforce section works here. State the question you cannot answer today, in one sentence.
"I do not know when the site goes down" is availability. "I cannot tell whether this is getting worse" is metrics. "I cannot reconstruct what happened at 04:12" is logs. "I know it is slow and not where" is tracing. "It is fine for us and users say otherwise" is front-end. "One office loses connectivity twice a week" is network.
Sentences containing two of those describe two problems. That is normal, and it is still worth solving them in order rather than buying a platform that claims all six.
Availability checking, as a class
- Best forAny team that needs to know before a customer tells them
- PricingFree self-hosted options exist; hosted tiers are the cheapest line in this market
- StandoutIt works when the rest of your monitoring is part of the outage
- Watch out forTells you that something is wrong and nothing about what, and it measures the network too
Metrics and time series, as a class
- Best forAlerting, trends, capacity, anything asked repeatedly
- PricingOpen source is free to license and costs storage and operations; hosted is billed per series or per host
- StandoutCheap per data point, fast over long spans, and the natural home for alert rules
- Watch out forCardinality decides the bill, and no metric can explain an individual request
Log management, as a class
- Best forReconstructing specific events and answering unanticipated questions
- PricingThe most expensive class at volume, in licensing and in storage alike
- StandoutAnswers questions nobody wrote a metric for, which is most questions during an incident
- Watch out forIndex cost dominates, and retention decisions here are where budgets actually go
Tracing and code-level tooling, as a class
- Best forSystems where one request crosses several services
- PricingUsually the premium tier of a commercial product, or self-hosted with real operational weight
- StandoutThe only class that answers where the time went across a boundary
- Watch out forRequires context propagated through everything, and sampling choices decide whether it helps during an incident
Front-end and real user measurement, as a class
- Best forAnyone whose users are on the far side of a browser or an app
- PricingCommonly bundled with a larger platform rather than bought alone
- StandoutSees the part of the experience that happens after your server is finished
- Watch out forBlind to the users who never loaded enough to report, which are the worst-affected ones
Network and device monitoring, as a class
- Best forEstates with physical hardware, offices, links and appliances
- PricingMature open source options exist; commercial products are priced per device
- StandoutSpeaks to equipment that cannot run an agent, which is most equipment
- Watch out forA separate discipline with its own vocabulary, and poorly served by application-oriented platforms
The platform question
Large commercial products cover four, five or six of these classes. That is genuinely convenient: one agent, one bill, one place to look, and correlation between classes that separate tools cannot easily do.
It is also the reason bills in this market become indefensible, since every class enabled is another metered stream, and the platform's weakest class is usually still enabled because it came with the others.
The question worth asking of a platform is not whether it covers a class but whether its coverage of the class you actually need is better than a tool that does only that. Frequently it is. Occasionally it is not, and the difference is worth an afternoon before a three-year contract.
What we cannot verify
No product is named on this page, and none of the general characterisations above should be read as describing any particular tool's implementation. Pricing structures differ between vendors and change; this site does not reproduce rates, and any figure quoted elsewhere here carries the date it was read. We have tested nothing and publish no scores. The situation guides in this section name specific tools, once the class question has been settled.
The short version
- Infrastructure monitoring is six markets, and products cover several at different depths.
- State the question you cannot answer today; the class names itself.
- Availability checking keeps working when the rest of your monitoring is in the outage.
- Metrics cannot explain one request; logs answer what nobody wrote a metric for.
- Tracing is the only class that answers where time went across a service boundary.
- A platform's convenience is real, and so is the metered stream behind each class.