The Logs That Matter Most: A Prioritized Source List
Every device on your network is willing to send you logs. The firewall will happily describe every packet it touches. The web server will narrate every image it serves. The printer will tell you, in detail, about its feelings. If you point all of that at one collector on day one, you will have a full disk and a slow search. You will have no better idea of whether anything bad is happening than you had before. The one line that mattered is buried under a million that did not.
Security log monitoring on a small team works only if you are ruthless about which sources you collect first. The good news is that the ranking is not mysterious. A handful of sources answer nearly every question you will ever need to ask during an incident. Those questions are: who logged in, from where, what did they touch, what did they talk to, and who changed what. Those sources are the ones to wire up before anything else.
This guide ranks them. For each source you will learn what it catches and roughly how much volume to expect. You will learn how much effort it takes to collect, and what to look at once it is flowing. Then comes the second tier, the list of things to deliberately skip for now, and a rollout order you can follow over a few weeks. The general method for deciding what any system should log — security or otherwise — lives in what to log; this guide is the security-specific answer.
In this guide
- How the ranking works
- Tier one: the six sources to collect first
- Tier two: worth adding once tier one is stable
- What to skip at first
- The priority table
- A rollout order
- Where to go next
How the ranking works
Three questions decide where a source lands. First, what does it catch? A source that reveals a stolen password outranks one that reveals a misconfigured print queue. Second, how much volume does it produce? Volume is cost — disk, search time, and your attention. A quiet source that says something important every time it speaks is worth more than a loud one that mostly says nothing. Third, how much effort does it take to collect? A source that already speaks syslog and needs one configuration line is the cheaper option. One that needs an agent, a parser, and a support ticket to the vendor costs more.
The tier-one list is the set of sources that score well on all three. Each one is also a join key for the others. Authentication logs tell you the username. The firewall and DNS tell you what that user's machine did. And the administrative logs tell you whether anyone changed the rules along the way. Together they cover the questions an investigation asks; individually, each is only a fragment.
Tier one: the six sources to collect first
1. Authentication logs: directory, VPN, and RADIUS
What it is: every place where a person or a service proves who they are. On a Windows network that means the domain controllers, which record every logon attempt against the directory. They record success, failure, the account name, the machine it came from, and the type of logon. Types include interactive at a keyboard, over the network, remote desktop, and so on. It also means the VPN concentrator, which records who connected from which outside address and for how long. Add any RADIUS server. Wireless access points and switches use that service to check a user's credentials before letting them onto the network.
What it catches: nearly everything that begins with a credential. Repeated failures followed by a success on one account. One account failing across many machines. Lockouts that nobody requested. A user who normally logs in from one desk suddenly logging in from a server. A VPN connection from a country the company has no business in. Because almost every intrusion involves using an account at some point, this source is the closest thing to a single most valuable feed.
Volume and effort: moderate volume and low-to-moderate effort. Domain controllers generate a steady stream of logon events, but far fewer lines than a firewall. Collection means enabling the relevant audit categories on the directory (they are not all on by default). It also means forwarding the security event log to your collector. VPN and RADIUS gear usually speaks syslog with a single configuration line.
2. Firewall logs: denies, admin events, and rule changes
What it is: the firewall records connection attempts it blocked (denies) and connections it permitted (allows). It also records administrative events, in a separate and often overlooked category. These are logins to the firewall itself, configuration saves, rule additions and deletions, and reboots.
What it catches: denies show you what is knocking on the door from outside and, far more interestingly, what is trying to leave from inside. An internal host repeatedly attempting connections to an unusual outside address on an unusual port is a workstation that wants investigating. Administrative events show you whether someone changed the firewall's rules. That is both a change-control question and a security one. A new rule that opens a port to the world at 11 PM on a Sunday is worth a conversation regardless of who added it. Turning the deny stream into something readable is its own craft, covered in deny log mining.
Volume and effort: effort is low — nearly every firewall speaks syslog. Volume is the catch: the allow stream can be enormous. Start with denies and administrative events only. Add allow logging later, selectively, for specific rules.
3. DNS query logs
What it is: a record of every name lookup your internal DNS servers answered: which client asked for which name, and what the answer was. DNS is the phone book of the network; before a machine can connect to anything by name, it has to ask the phone book first.
What it catches: the name before the connection. A firewall log tells you a workstation connected to an address. The DNS log tells you what name it asked for to get that address, which is usually far more meaningful to a human. Malicious software very often relies on DNS to find its way home. A workstation may look up a long list of random-looking names, or the same odd name every few minutes all night. That stands out clearly in a DNS log and nowhere else. DNS logs also answer the investigator's question "what else looked up that name?" — which is how one infected machine leads you to the other three.
Volume and effort: volume is high — every application on every machine generates lookups — but the lines are short and compress well. Effort is moderate: query logging must usually be switched on deliberately. On Windows DNS servers it goes to a dedicated log or file that you then forward. Log queries only, not the server's internal debugging, and the volume becomes manageable.
4. Endpoint protection alerts and disablement events
What it is: whatever anti-malware or endpoint protection product runs on your workstations and servers produces two kinds of event worth collecting. The first is the detection: something was found, on which host, and what was done about it. The second, and easily the more important, is the status event. Protection was disabled, real-time scanning was turned off, definitions are badly out of date, or the agent stopped reporting entirely.
What it catches: detections are obvious and useful, but the disablement events are the real prize. Protection does not switch itself off. When it does, an administrator may have done it for a reason that should be in a ticket. Or someone with administrator rights on that host did it for a reason that should be in an incident report. Either way, you want to know within minutes, not at the next quarterly review.
Volume and effort: very low volume — on a healthy network this source is nearly silent, which is exactly why every line deserves a look. Effort varies: most products have a central console that can forward events by syslog or email. Even the email route is acceptable for a source this quiet.
5. Administrative actions: accounts, groups, privileges, configuration
What it is: the record of changes to the things that control access. On the directory, this includes new accounts created, accounts enabled or disabled, password resets, and users added to or removed from groups. Any change to the groups that grant administrative rights especially matters. On servers and network gear: configuration changes, new scheduled tasks or services, and changes to who can log in. Much of this lives in the same security log as your authentication events. But it is worth treating as its own source because it is watched differently. Authentication is watched for patterns; administrative actions are watched line by line.
What it catches: the moment an intruder — or a well-meaning colleague — grants themselves more power. A new account that nobody requested. A user quietly added to an administrators group. A service account whose password was reset outside a change window. These are rare, high-value events, and on a small network you can reasonably read every single one.
Volume and effort: low volume, low effort once directory auditing is enabled for the authentication source above. The additional work is making sure the account-management and group-membership audit categories are on. Forward configuration-change events from the firewall and switches, too. These usually arrive with the syslog stream you already set up.
6. Email gateway events
What it is: whatever sits between the internet and your users' mailboxes records what it accepted, what it quarantined, and why. It might be a hosted filtering service, an on-premises gateway, or the built-in protection of a hosted mail platform. It also records administrative changes: new forwarding rules, new mailbox delegates, and changes to filtering policy.
What it catches: email is the front door for the majority of what goes wrong on a small network. The gateway's quarantine log therefore tells you what kind of week you are having and which users are being targeted. The administrative events are more pointed. A mailbox may suddenly forward every message to an outside address. That is one of the most common signs that its owner's password is no longer only theirs. That single event type justifies collecting the source.
Volume and effort: moderate volume, moderate effort. Hosted services typically offer an export or a reporting interface rather than syslog, so this source often needs a small scheduled script rather than a listener. Start with quarantine summaries and administrative changes; do not try to log every accepted message.
Tier two: worth adding once tier one is stable
These sources are genuinely useful, but each one comes with a trade-off. It may answer a narrower question or produce more volume than its value justifies at first. Or it may take more effort than a small team should spend before tier one is quiet and trusted.
- Web proxy or content-filter logs. Which user's machine visited which site. Excellent for investigation ("what did that workstation browse to right before the detection?"), heavy in volume, and only available if you have a proxy in the path. Add it when you have a specific question it would answer.
- DHCP lease logs. Which device held which IP address, and when. This is the source that turns "10.20.4.17 did something" into "the laptop with this hardware address did something," which matters because addresses move. Low volume, low effort, and mostly used to support the other sources rather than to detect anything on its own.
- Switch port events. Links going up and down, and — if you use port security or 802.1X — devices being refused. Useful for spotting an unknown device plugged into a conference room, noisy on switches with flapping ports, and easy to collect since switches speak syslog.
- Backup job results. These are not a security source in the classic sense. But a backup that starts failing right before an incident, or a backup set that someone deleted, is very much a security event. It is one line a day per job, so the cost of collecting it is nearly nothing.
What to skip at first
Discipline about what to leave out is what makes tier one work. Skip, for now, application debug logs: the chatty internal narration of web servers, databases, and line-of-business applications. They are invaluable when troubleshooting that application and nearly useless for security until you have a specific reason to look. Skip firewall allow logs at full volume. As noted above, they can dwarf everything else combined. The questions they answer are better answered by DNS logs at a fraction of the size. Skip workstation event logs in bulk. Forwarding every event from every desktop is a project in itself. The security-relevant subset (logons, protection status) is already covered by the directory and the endpoint product. And skip anything from devices you would not act on: the printer, the smart TV in the lobby, the test lab. If an alert from that source fired tomorrow morning and you would not do anything, it is not a source yet. It is future noise.
Tip: "skip at first" does not mean "never." Every source above can be added later, and some will earn their place the first time an investigation stalls for lack of them. Keep a short list titled "sources we wished we had" and revisit it after each incident review.
The priority table
| Rank | Source | What it catches | Volume | Effort |
|---|---|---|---|---|
| 1 | Authentication (directory, VPN, RADIUS) | Guessed or stolen credentials, lockouts, logins from odd places, service accounts used by people | Moderate | Low–moderate |
| 2 | Firewall (denies, admin, rule changes) | Internal hosts reaching out oddly, external probing, unauthorized rule changes | Moderate (denies only) | Low |
| 3 | DNS queries | Names looked up before connections; beaconing; which other hosts asked for the same name | High, but compact | Moderate |
| 4 | Endpoint protection | Detections; protection disabled or agent silent | Very low | Low–moderate |
| 5 | Administrative actions | New accounts, group changes, privilege grants, config changes | Low | Low (with #1 enabled) |
| 6 | Email gateway | Targeted users, quarantines, forwarding rules and delegates | Moderate | Moderate |
| Tier 2 | Web proxy | Browsing history for investigations | High | Moderate |
| Tier 2 | DHCP leases | Which device held which address, and when | Low | Low |
| Tier 2 | Switch port events | Unknown devices, refused connections | Low–moderate | Low |
| Tier 2 | Backup job results | Failed or deleted backups before or during an incident | Very low | Low |
A rollout order
You do not need to wire up all six sources in one weekend. And you should not try, because each new source needs a week or two of watching before you know what its normal looks like. A sequence that works:
- Stand up the collector and point the domain controllers at it. Enable the logon, account-management, and group-membership audit categories on the directory. Confirm events arrive, and confirm timestamps agree with the collector's clock.
- Add the firewall, denies and administrative events only. One syslog line in the firewall's configuration. Confirm you can see a deny and a configuration save in the collector within a minute of each happening.
- Add the VPN and RADIUS. Same mechanism. Now every way a person gets onto the network is recorded in one place.
- Add endpoint protection. Configure the console to forward detections and status changes. Test by pausing protection on a lab machine and confirming the event arrives.
- Turn on DNS query logging. Watch the volume for a few days and check disk consumption before you decide on retention for this source.
- Add the email gateway. Usually a scheduled export script. Start with administrative changes and quarantine summaries.
- Only now, start writing detection rules — the starter set is in detections a small team can actually run.
Keep a plain text inventory of what you have connected, because in six months you will not remember, and neither will whoever inherits it. Something as simple as this is enough:
# source-inventory.txt # source host/service method enabled retention owner directory-auth dc01, dc02 event fwd yes 1 year sam firewall-deny fw-edge syslog yes 6 months sam firewall-admin fw-edge syslog yes 1 year sam vpn vpn-gw syslog yes 1 year sam radius nps01 event fwd yes 6 months sam endpoint-protect epp-console syslog yes 1 year sam dns-query dc01, dc02 file copy yes 90 days sam email-gateway hosted export script yes 6 months sam web-proxy - - not yet - - dhcp dc01 file copy not yet - -
Two columns in that file do more work than they look like they do. Enabled is what you check every week: a source that silently stops sending is a blind spot. A blind spot that appears right before an incident is rarely a coincidence. Retention is what you will be asked for by an auditor or an insurer. Having the answer written down — per source, in a file you can show them — ends that conversation in one sentence.
Where to go next
Six sources, wired up in order, watched for a couple of weeks each, and written down. That is the foundation, and everything else in Security Log Monitoring on a Budget (SIEM-lite) stands on it. The detection rules assume these sources exist. The correlation guide joins them by user and host and time. The guide to outgrowing do-it-yourself asks whether this list has become too long for scripts to handle. If you are still deciding whether any of this is worth doing, start with what a SIEM does, which explains the five jobs these sources feed.
Put this into practice with SysPeep Monitor
Every source on this list is a machine or service that can quietly stop sending. A source that goes silent is a blind spot you will not notice until you need it. SysPeep Monitor keeps watch on the domain controllers, firewall, DNS servers, and collector behind your log pipeline over SNMP and WMI. It alerts when one goes down or runs out of disk. It gives you the health history to sit alongside the logs. Try the free trial on any Windows machine.
Download Free TrialLearn About SysPeep