Egress Monitoring on a Budget
You have read that you should be watching data leave your network. You have also read the brochures. The brochures want a budget you do not have, an agent on every endpoint, and a six-month "classification project" before anything useful happens. Meanwhile the question from management — "would we know if our data walked out?" — is still sitting in your inbox. The honest answer is still "probably not."
Here is the part the brochures leave out. Most of what you need to answer that question is already on your network, already producing data, and already free. Your firewall knows every outbound session it allowed and how many bytes went through it. Your routers and switches can summarize every conversation they carry. Your DNS servers know which names your hosts asked about right before they connected somewhere. Nobody sells you those three; you just have to turn them on, point them at a collector, and ask them the right questions.
Egress monitoring means watching traffic that leaves your network for the outside world — the outbound half of every conversation. This guide shows how to do it with what you have. It covers what each of the three free sources gives you and the three reports worth building from them with nothing more than scripts. It covers what commercial data-loss-prevention tooling adds if you ever get the budget. And — because honesty matters more than reassurance — it covers exactly what the free approach cannot see.
In this guide
- The free trio
- Joining the three sources
- Three reports you can build with scripts
- What commercial DLP adds, and what it costs in effort
- The honest limits
- What the free trio can and cannot see
- Starting small: the ten most sensitive hosts
- Where to go next
The free trio
Each of the three sources answers a different piece of the same question. None is sufficient alone, and together they cover most of what a defender needs to see outbound patterns.
Firewall logs: allowed outbound sessions with byte counts
Nearly every firewall can log the sessions it handles. Most are configured to log only what they block, because that is what people think of as "security events." For egress monitoring you need the opposite: log what was allowed outbound. Make sure the log entry written when the session ends includes the byte counts in each direction. That record includes internal address, external address, port, protocol, start time, duration, bytes sent, and bytes received. It is the single most useful line of text on your network for this job. Send it to a central collector over syslog rather than leaving it on the firewall, where it will be overwritten within hours. The costs are disk space on the collector and a little CPU on the firewall, and both are small compared with the value.
Flow data: per-host, per-destination volume
Flow data is the family of summaries that routers and switches export about the conversations passing through them. The summaries cover who talked to whom, on which ports, for how long, and how many bytes and packets. Flow data overlaps with the firewall's session log at the perimeter, but it adds two things. It sees traffic the firewall does not, such as internal host-to-server movement — useful for spotting the "big internal pull, then big external push" pattern. And it is exported by devices you already own, so a second vantage point is often free. The Flow Monitoring guides cover setting up export and reading the reports. For egress work, export from the device closest to your internet edge first. Be aware that sampled flow can under-count small transfers.
DNS logs: the names behind the destinations
Firewall and flow records identify destinations by IP address, and an address on its own tells you little. Your internal DNS resolvers — the servers that turn names into addresses for every host on the network — can log each query. Such a log records which host asked for which name, when. Joined with the session data, that gives you the name a host looked up moments before it connected somewhere. That is often the difference between "a cloud service we use" and "a name nobody has heard of." It also reveals the opposite case: a large session to an address that no host ever looked up. That means something connected directly to a raw IP without asking DNS at all. That is unusual for ordinary software and worth a flag on its own. DNS can also be abused as a transport in its own right. That is a different problem with its own detection approach, covered in the DNS Threat Sensor guides.
Joining the three sources
The three sources become powerful when joined, and the join is simple: internal host plus time. A session in the firewall log at a given moment from a given host matches a flow record at the same moment from the same host. The flow record also matches a DNS query from that host a second or two earlier. You do not need a database engineer to do this. A daily script that reads all three, normalizes timestamps to one reference, and groups by host is enough for an organization of a few hundred devices. Here is what each contributes to a single joined record:
| Question | Firewall log | Flow data | DNS log |
|---|---|---|---|
| Which host sent it? | Yes (internal address) | Yes | Yes (the asker) |
| Where did it go? | Yes (address, port) | Yes (address, port) | The name that resolved to it |
| How much went out? | Yes, at session end | Yes, per flow | No |
| When, and for how long? | Yes | Yes | When the name was asked |
| Was there an internal gather first? | Rarely (internal traffic bypasses it) | Yes, if the internal path exports flow | Sometimes (internal name lookups) |
| Did the host skip DNS entirely? | Only by absence in the DNS log | Same | Yes: no query, yet a session followed |
One practical trap: log at a point where you can still see the real internal address. If your only outbound log comes from a device after address translation has already happened, every session appears to come from the firewall itself. In that case, the per-host view is gone. Log on the inside interface, or from the device that performs the translation, so the original host is preserved.
Three reports you can build with scripts
You do not need dashboards to start. You need three plain reports, produced every morning, short enough to read with coffee. Each corresponds to one of the exfiltration dimensions described in What Data Exfiltration Looks Like, and each compares today against the profiles from Outbound Baselines.
Report 1: top uploaders by host
Per host, total bytes out yesterday, ranked, with the host's role and its typical daily figure beside it. The ranking on its own is dominated by the usual suspects — backup targets, cameras, the mail relay — which is fine. The column that matters is the ratio to the host's own normal. Sort by ratio and the interesting hosts float to the top regardless of absolute size.
Report 2: new destinations
List every host-and-destination pair seen yesterday that was not in the running table before. Include the DNS name if one was looked up, the owning network, and the bytes sent to it. Flag pairs that are new for the whole network, and flag any that received more than a modest amount on their first appearance. After the first few weeks this report is short on most days, which is exactly the point.
Report 3: after-hours uploads
Per host, report bytes out during the host's off-hours window, excluding destinations covered by your allowlist of legitimate night jobs. Include the top destination and whether it was first-seen. The allowlist and the rule logic behind this one are described in Odd-Hours Transfers. The report is just the daily, human-readable form of the same rule.
Laid out on one page, the morning email looks roughly like this:
DAILY EGRESS REPORT -- yesterday, all times local [1] TOP UPLOADERS BY RATIO TO OWN BASELINE host role out_yday typ_out ratio top_dst (share) WS-FIN-04 workstation 3.9 GB 310 MB 12.6x 203.0.113.58 (97%) *NEW* WS-ENG-22 workstation 2.1 GB 900 MB 2.3x code-host (88%) BACKUP-02 backup 181 GB 170 GB 1.1x offsite-tgt (99%) [2] NEW DESTINATIONS (host, dst) pairs not seen before host dst name_looked_up owner_net bytes_out scope WS-FIN-04 203.0.113.58 (none -- direct to IP) hosting-prov 3.8 GB NETWORK-NEW WS-MKT-09 198.51.100.20 cdn-of-known-vendor known-good 40 KB host-new LAPTOP-HR-03 192.0.2.77 webinar-platform known-neutral 6 MB host-new [3] AFTER-HOURS UPLOADS (not allowlisted, above per-role floor) host window bytes_out floor top_dst first_seen WS-FIN-04 02:41-03:25 3.8 GB 120 MB 203.0.113.58 yes (network) FILES-01 23:00-23:40 210 MB 4 GB (below floor; shown for context)
The same host appears in all three sections, which is how a report earns its keep. One line in one section is a curiosity. The same host in all three is the start of a case. Building the pipeline that produces this is a short numbered project:
- 1. Collect. Send firewall session logs to a syslog collector. Send flow exports to a flow collector (a small host will do). Send DNS query logs to the same place or a shared file store.
- 2. Normalize. One script per source that emits the same simple record: timestamp, host, destination, port, bytes out, bytes in, name if known.
- 3. Aggregate daily. Sum per host, per host-and-destination, and per host per window; count distinct destinations.
- 4. Diff. Compare today's host-destination pairs against the running table; append new ones with a first-seen timestamp.
- 5. Compare and rank. Join with the baseline table; compute ratios; apply the allowlist to the after-hours section.
- 6. Send. Email the three sections to yourself every morning. Keep the daily outputs; they are evidence later.
This is a weekend of scripting for someone comfortable with a shell or a scripting language, and it needs no product at all. If you later adopt a log platform or a flow tool, the same three reports move into it unchanged.
Tip: keep the retention of the raw daily records longer than feels necessary — many months, not weeks. The slow-leak pattern, a small amount every day to the same place, is invisible in any single day's report. It only appears when you sum a long history per destination. Disk is cheaper than the conversation where you explain that the records were rotated away.
What commercial DLP adds, and what it costs in effort
Data loss prevention (DLP) is the generic name for tooling designed to stop sensitive data leaving, rather than to notice that data left. Vendors differ, but the category adds three things the free trio structurally cannot:
- Content awareness. This means inspecting what is inside a transfer. It means recognizing a document as a contract, a spreadsheet as customer records, a string as a card number. Usually, it works by classifying files in advance and by pattern-matching content at the point where it can still be read. That is typically on the endpoint itself or inside a sanctioned cloud service.
- Endpoint controls. An agent on each device can see removable media, printing, clipboard and screenshot activity, and uploads through the browser. It can block them according to policy. This is the only way to see the channels the network never sees.
- Enforcement. Blocking, quarantining, or warning the user at the moment of the attempt, rather than reporting it the next morning.
The price on the quote is the smaller cost. The larger cost is effort. Someone has to decide what counts as sensitive, then label it or teach the tool to recognize it. They have to write policies that block the right things without blocking payroll. They have to deploy and maintain agents on every endpoint and then live with the tuning period. During that period, the tool stops a great many legitimate things and users learn to route around it. Organizations that do this well tend to have first learned, from cheap egress reporting, which hosts and which data actually matter. The free trio is not a lesser version of DLP. It is the homework that makes DLP a sensible purchase rather than an expensive guess.
The honest limits
Now the part you need to be able to say out loud to management. The free approach sees patterns in the envelope — who, where, when, how much — and nothing else. Specifically:
Encrypted channels hide content, not shape. Almost everything outbound is encrypted, so you will never see filenames or file contents in these logs. You will still see volume, destination, timing, and protocol, and those are enough to notice most of the patterns in this pillar. The limit is that you cannot tell a large upload of public brochures from a large upload of the customer list. That judgement comes from context and people, not from the network.
Removable media and printing are invisible. A file copied to a USB stick or printed and carried out never touches the perimeter. The network has no view of it at all. Only endpoint controls or physical policy address this. It is worth being explicit about that gap rather than letting anyone assume the network covers it.
Cloud-to-cloud movement bypasses you. Suppose your data lives in a cloud service and someone shares it from that service to an outside account. In that case, the transfer happens between two data centres and never crosses your network. The audit logs of the cloud service itself are the only place that movement is recorded.
Personal devices and other networks. A phone photographing a screen, or a laptop on a cellular hotspot, produces no perimeter record. Mobile connections and guest networks you do not log are simply outside the picture.
Sampling and blind spots inside. Sampled flow may miss small transfers. A segment that does not export flow leaves the internal "gather" half of a staging pattern unseen. A site with its own internet breakout that is not logged is a hole. Know where your holes are and write them down. A list of known blind spots is more honest, and more useful in an incident, than a vague sense of coverage.
What the free trio can and cannot see
| Channel or fact | Firewall + flow + DNS | Notes |
|---|---|---|
| Outbound volume per host, per destination, per hour | Yes | The core of every report in this guide. |
| First-seen destinations and the names behind them | Yes | DNS supplies the names; absence of a lookup is itself a signal. |
| Transfers at odd hours | Yes | Needs the working window from the baseline. |
| A new protocol or port on a host | Yes | Visible even when the session is encrypted. |
| Internal gather before an external push | Partly | Only where internal paths export flow. |
| What was inside the transfer | No | Encryption hides content; only endpoint or cloud-side inspection can see it. |
| Which user was at the keyboard | Indirectly | Host to user mapping from your directory or login logs; not from the network. |
| USB, printing, photos of screens | No | Outside the network's view entirely. |
| Sharing from one cloud tenant to another | No | Never crosses your perimeter; use the service's own audit log. |
| Traffic on unlogged breakouts or cellular | No | A known blind spot to list, not a surprise. |
Read the table as a coverage map, not as a verdict. A defender with the top five rows working reliably is far ahead of one with nothing. The bottom five rows are the honest agenda for what a budget, if it ever comes, should buy.
Starting small: the ten most sensitive hosts
Do not try to log everything on day one. Pick roughly ten hosts and do the whole job properly for them. That means outbound firewall logging with byte counts, flow from the nearest exporter, and DNS query logging for their addresses. It also means a baseline row for each and all three morning reports scoped to just those hosts. Choose them by what they can reach, not by how important the machine itself seems. Examples are the file server holding contracts and the database behind the customer system. Others are the workstations of people with access to payroll, HR, and the executive mailbox. Also choose any host that already has a legitimate reason to talk to the outside a lot. Those are the easiest places for a bad transfer to hide.
Ten hosts keep the reports short enough to actually read, which means you will read them, which is the entire difference between monitoring and decoration. After a month you will know what the reports look like on a boring day. By then, your allowlist and baseline rows will be trustworthy. Adding the next fifty hosts is mostly a matter of copying what already works. You may someday need the more general skill of finding a single host that is eating bandwidth — a capacity problem rather than a security one. In that case, the top-talker workflow covers that separately.
Where to go next
You now have a plan that costs nothing but time. It uses three sources you already own, joined on host and time. It has three morning reports that map onto volume, destination, and hour. It includes an honest list of what the approach cannot see and a starting scope small enough to succeed. The reports will, sooner or later, put the same host in all three sections on the same morning. When that happens, Investigating a Suspected Exfiltration is the calm, ordered way to handle it. The rest of this series lives at Spotting Data-Exfiltration Patterns.
Put this into practice with SysPeep Monitor
The free trio only works while the collectors behind it stay healthy. A syslog server with a full disk or a flow collector that quietly stopped is a blind spot you will discover at the worst time. SysPeep Monitor watches those hosts alongside everything else it discovers. It provides disk, service, and reachability checks over WMI and SNMP, plus interface traffic graphs on the internet edge. It sends alerts by email when a collector goes quiet. All of this comes from one web console. The free trial installs on any Windows machine.
Download Free TrialLearn About SysPeep