LDAP and Active Directory Health for Network Admins
Here is an uncomfortable fact about the org's most important service: when it gets sick, you get blamed. Active Directory — the directory of every account, group, computer, and policy you have — sits behind every login. When logins misbehave, the first diagnosis is always "the network is being weird." So the network admin ends up defending the network's honor in incidents that are really directory incidents. Usually, that happens without a single check that could prove it either way.
Let us be honest about scope up front: this is not a course in administering domain controllers. Directory administration — schema, sites, operations roles, forest design — is its own deep specialty. This guide will not pretend to teach it in three thousand words. What it teaches instead is the part that rightfully belongs to you: the monitorable surface. From the network seat you can watch whether every domain controller is reachable and whether the directory answers real questions quickly. You can watch whether the controllers agree with each other. And you can watch whether the two services AD is entangled with — names and time — are holding it up or dragging it down.
That surface turns out to be most of what matters day to day. By the end of this guide you will have a concrete check list. It covers what to watch on every DC and the one synthetic test that honestly answers "is the directory okay?" It covers what replication health means and how to keep an eye on it without a specialist's toolkit. And it covers which signals deserve a page versus a spot on the weekly review.
In this guide
- The directory in plain language
- Reaching every DC (yes, all of them)
- The honest check: a synthetic LDAP bind
- Replication: the controllers must agree
- The entanglements: DNS and time
- What to alert on, what to review weekly
- One DC down is not an outage — model it that way
- Where to go next
The directory in plain language
Think of the directory as the organization's records office. It holds the file on everyone and everything. That includes user accounts and their password verifiers, group memberships, computers, printers, and the policies that say who may do what. Every login in the building ultimately walks up to this records office and asks it to vouch for someone. That includes logins to a PC, a file share, an application, the VPN by way of RADIUS.
Three names cover the moving parts. Active Directory (AD) is Microsoft's implementation of the records office, and the one most orgs run. LDAP (the Lightweight Directory Access Protocol) is the standard language for asking a directory questions — "what groups is dana in?" LDAP is spoken by AD and by non-Microsoft directories alike. That is why this guide's checks work either way. Kerberos is the authentication protocol that does the actual vouching in a Windows domain, using time-stamped tickets. And the records office is not one building. It is replicated across servers called domain controllers (DCs). Each holds a full copy, and each is able to accept changes. All constantly synchronize so they tell one story. Hold onto that image — several offices, one set of records — because both the resilience and the strangest failures of AD come from it.
If you are wondering why the network admin should care about any of this beyond logins: the directory is also infrastructure's own address book. Your RADIUS servers query it for every Wi-Fi and VPN authentication. Applications bind to it to check group membership. Printers and scanners look up users in it for scan-to-email. Monitoring tools authenticate their consoles against it. The directory is not one service among many on your network. It is the service most of the others phone before doing anything interesting. That makes its health inseparable from the health of everything you already watch.
Reaching every DC (yes, all of them)
The first layer of monitoring is bluntly simple: is every domain controller up and reachable? The emphasis belongs on every. Clients find DCs through lookups that rotate and fail over. That is resilience working as designed. It means a dead DC does not necessarily announce itself. Users limp along on the survivors. Logins get slower or intermittently weird, and the corpse can lie undiscovered for days. Checking "the domain" is like phoning any branch of the records office and concluding all branches are fine because someone answered. Monitor each DC by its own name and address, individually.
Branch sites raise the stakes on this. Orgs commonly place a DC at each significant site so logins stay local and survive WAN outages. That means the branch DC is the only records office that matters to the people in that building. Monitor it from two perspectives if you can. Watch from the central monitoring server (proving the WAN path and the DC). Also watch from a probe inside the branch (proving what branch users actually experience). A branch DC that is fine from headquarters and unreachable from its own LAN is a switch or VLAN story. Knowing that instantly is exactly the kind of triage this whole pillar is about.
Reachability is more than a ping, because a DC is really a bundle of services on one box, each on its own port. You do not need to memorize the full list. But knowing the port set generically turns "the DC is up but something is off" into "this specific service is not answering." The short version:
| Port (typical) | What it serves | What breaks without it |
|---|---|---|
| 53 | DNS — most DCs also answer name lookups | Clients cannot find the domain at all |
| 88 | Kerberos authentication | Logins fail or fall back insecurely |
| 389 / 636 | LDAP / LDAP over TLS | Directory queries fail — apps, RADIUS, everything |
| 445 | File service — policy scripts and logon shares | Group policy stops applying cleanly |
| 3268 / 3269 | Global catalog — wider, cross-domain lookups | Address books and cross-domain logins misfire |
A good monitoring setup checks the ports that matter to your environment on every DC. That way, "DC3 is up but its LDAP port stopped answering" becomes a distinct, actionable alert instead of a mystery distributed across tomorrow's helpdesk queue.
The honest check: a synthetic LDAP bind
Ports answering still only proves that doors are open, not that anyone useful is behind them. The check that honestly answers "is the directory working?" is the same gold standard used across all core services: a synthetic transaction. For a directory, that means an LDAP bind — the protocol's word for logging in — followed by a small query. A dedicated monitoring account performs the bind-and-query against every DC, on a schedule, with the response time recorded.
# Bind to ONE SPECIFIC DC and look up the probe account itself ldapsearch -H ldaps://dc2.corp.example \ -D "svc-ldapprobe@corp.example" -w PASSWORD \ -b "dc=corp,dc=example" "(sAMAccountName=svc-ldapprobe)" dn # Success: a dn line comes back within your threshold # Failure flavors: no route / refused (port), invalid credentials # (account or its password), timeout (DC grinding), no results # (query or base wrong — fix the check, not the DC)
One passing bind-and-query proves a remarkable amount in one go. The DC is up, and its directory service is listening. It accepted a real credential and searched its database. And it did all of that within your time limit. Run it against each DC every few minutes and graph the response time per controller. Healthy answers come back in tens of milliseconds on a LAN, maybe low hundreds across a WAN link. Your baseline will declare itself within days. The graph is where trouble whispers first. A DC whose bind time quietly triples is telling you about its disk, its memory, or its overload before any user does. Every one to five minutes is a sensible cadence. Faster buys little, and slower can miss the short brownouts that make logins feel haunted.
Treat the probe account like the small security liability it is. Give it minimal rights (it only needs to read, and only a little), an obvious name, and a documented owner. Have a deliberate answer for password expiry. And prefer the TLS-wrapped port for the check. It is what your applications should be using, so it is what your monitoring should be proving. As a bonus, checking the secured port quietly puts the DC's certificate in your line of sight. An expiring one is a directory outage on a timer.
Tip: point one extra synthetic bind at the directory the way clients actually find it — via the domain name rather than a specific DC. The per-DC checks tell you which controller is sick. The domain-path check tells you what users are experiencing. When the per-DC checks disagree with the domain-path check, the problem usually lives in the finding. That is a name-resolution story, covered below.
Replication: the controllers must agree
Several offices, one set of records — but only if the offices keep synchronizing. AD is multi-master. A change (new user, password reset, group edit) can be written to almost any DC. That DC then replicates the change to its partners on a schedule of minutes. When replication falls behind or fails, the offices start telling different stories. The symptoms are among the strangest in IT. A password reset works at headquarters but fails at the branch. A user added to a group can open the share from one PC but not another. An account you unlocked re-locks itself, because a DC that never heard about the fix keeps re-asserting the lock.
The deep diagnosis of replication belongs to directory tooling. The directory ships with built-in commands that print a per-partner summary of what replicated when, what failed, and how long any partner has gone unheard from. It also ships with a health-check utility that runs a battery of self-tests on a DC. You do not need to be a specialist to use them at the summary level. Run the replication summary, read the failure count column, and note the "longest time since last success" figure. The tools' output is exactly the evidence a directory specialist or vendor will want if escalation is needed.
Here is what reading the summary looks like in practice, generically. You run the replication summary on any DC and get a table. There is one row per replication partner, with columns for how many attempts failed and how long since the last success. The healthy reading is glorious boredom — zeros in the failure column, minutes in the last-success column. The reading that earns a calendar entry is a partner showing accumulating failures, or a last-success measured in days. You do not need to interpret the error codes yourself. The row tells you which two DCs cannot talk and since when. That is nine-tenths of the diagnosis. And very often the "since when" lines up suspiciously with a firewall change or a WAN hiccup already in your change log.
Two things put replication legitimately on the network admin's desk. First, replication rides on your network. A flaky WAN link, an overzealous firewall rule, or a routing change between sites surfaces as replication failure. You will diagnose it faster than a directory admin staring at directory logs. Second, replication health degrades silently — nothing user-visible happens on day one. That makes it a perfect weekly-review item. Take a two-minute glance at the summary, looking for anything nonzero in the failure column or any partner silent for longer than a day. Silent-for-weeks replication partners are how orgs end up with a "zombie DC" whose stale copy of reality bites months later.
The entanglements: DNS and time
Two outside services are so load-bearing for AD that their failures are functionally directory failures, and any honest health picture has to include them.
Names. Active Directory lives and dies by its DNS records. Clients discover which servers are domain controllers by looking up special service records the DCs publish. So when those lookups fail, a perfectly healthy domain becomes unfindable and the whole org appears logged out of existence. Name resolution has its own full pillar of guides. Everything there applies double wherever AD is involved.
Time. Kerberos tickets are time-stamped and validated against a skew tolerance — commonly five minutes by default. So a drifted clock turns correct passwords into rejected logins, one machine at a time, in the most random-looking pattern imaginable. The domain normally syncs itself, but "normally" is a hypothesis your monitoring should test. The companion guide on NTP and clock drift covers watching offset properly. For this guide's purposes, remember one rule: clock-offset checks on every DC are directory monitoring.
What to alert on, what to review weekly
The directory generates enough observable signals to bury you, so the craft is sorting them into "wake me" and "tell me Friday." A workable split:
| Signal | Alert now | Weekly review |
|---|---|---|
| Any DC unreachable, or a core port not answering | Yes | — |
| Synthetic bind fails on any DC (two consecutive) | Yes | — |
| Bind latency far above baseline on any DC | Yes | Trend line |
| Domain-path check failing (users' route to the directory) | Yes, loudly | — |
| DC clock offset past threshold | Yes | Drift trends |
| Directory certificate nearing expiry | Warn weeks ahead | Inventory check |
| Replication summary: failures or silent partners | Only if user-visible | Yes, every week |
| Account lockout volume, auth error trends | — | Yes |
The weekly column is not the junior varsity. Replication drift and slowly climbing bind times are exactly the problems that never page you and then define a terrible month. Fifteen minutes on a Friday, same checklist every time, is the cheapest insurance the directory sells.
Expect to tune. The first week of directory monitoring always surfaces surprises. One example is a DC that reboots nightly for updates and pages you at 3 a.m. (add a maintenance window). Another example is a WAN site whose bind latency is legitimately triple headquarters' (give it its own baseline). Another example is a backup job that makes one DC crawl for twenty minutes (either accept it with a scheduled exception or fix the job). None of that is failure. It is the monitoring learning your environment. The rule from every other corner of this library applies here with extra force, because the directory is the loudest thing you will ever monitor. An alert you routinely ignore is worse than none.
Tip: watch account lockout volume as a weekly trend even if you never alert on it. A slow rise in lockouts usually means a stale saved credential grinding away somewhere — a service, a script, a forgotten phone. Finding it on your terms beats finding it the week it locks out an executive during a board meeting.
One DC down is not an outage — model it that way
Redundancy makes directory alerting philosophically interesting. With three DCs, one dying is not a login outage — users fail over and barely notice. But it is absolutely not "fine," because you are now one failure from a very bad day, and possibly already serving slower logins at one site. Your monitoring should say exactly that. Model the directory as a service whose health is derived from its members. One DC down means "directory degraded — investigate today." The last DC serving a site down means "site auth outage — investigate now." All DCs failing the synthetic bind means the org is off the air. Most monitoring systems can express this with dependency and cluster-style rules. The effort pays off twice — calmer paging, and alerts whose wording matches reality. The craft of building those service-level dependency models is covered in the guide to dependency maps that tame alert storms.
However you model it, resist the one genuinely wrong setting: treating each DC as an independent, equally urgent, page-at-3-a.m. alert. That configuration pages you hardest precisely when the system is doing its job of surviving. And alert fatigue against your most important service is how the real outage, months later, gets snoozed.
Where to go next
You now hold the network admin's honest share of directory health. Every DC is watched individually, and a synthetic bind proves real answers at real speed. Replication is reviewed weekly at the summary level. Clocks and names are held to account, and alerts distinguish degraded from down. The directory rarely suffers alone. It is the records office behind the RADIUS gatekeeper. Both sit below the waterline of the dependency iceberg. When logins break anyway, the authentication-path playbook walks the whole chain in order. The rest of the family lives at Core Services: NTP, RADIUS, LDAP/AD.
Put this into practice with SysPeep Monitor
Domain controllers are Windows servers. Watching them is home turf for SysPeep Monitor. Auto-discovery finds every DC. WMI and SNMP collectors track their health and resources. Dependency-aware alert rules can tell "one DC down, degraded" from "authentication outage". So the page you get matches the problem you have.
Download Free TrialLearn About SysPeep