The server monitoring underneath the AI bottleneck analysis.
WPLoadTester's Server Monitoring agent runs during every load test. Free captures CPU, memory, and disk on every monitored host. Pro adds network, TCP, and process metrics, plus integration with AWS CloudWatch and Dynatrace. The Analytics Dashboard plots all of them, auto-flags any resource that correlates above 0.94 with response-time degradation, and on Pro the AI Bottleneck Report names the saturated resource in plain English.
Free captures the foundationals. Pro captures everything.
The agent collects metrics automatically on every monitored host. There is no metric-selection step to configure and no counter list to maintain.
Free edition. Foundational metrics, three groups.
- CPU: CPU Utilization, CPU Steal Time, Context Switches.
- Memory: Memory Utilization, Page Ins, Page Outs, Major Page Faults, Cache Memory Allocation.
- Disk: Disk Reads, Disk Writes, Disk I/O Time, Average Disk Service Time, Disk Queue Length.
CPU, memory, and disk cover most bottleneck stories. If your test server pegs CPU at 97 percent at the same user level response time goes vertical, the Free edition surfaces that without you needing Pro.
Pro edition. The full 24-metric set, plus integrations.
- Network: Bandwidth In, Bandwidth Out, Network Packets Received, Network Packets Sent, Network Packet Errors (Rx), Network Packet Errors (Tx), Network Collisions.
- TCP: TCP Connections Established, TCP Connection Failures, TCP Segments Retransmitted.
- Processes: Process Queue Length.
- AWS CloudWatch integration: existing CloudWatch metrics pulled into the Server Performance overview alongside the load-test data, on the same x-axis as response time.
- Dynatrace integration: same idea for teams already running Dynatrace.
Network and TCP metrics matter the moment a bottleneck is anywhere other than the box itself: load balancers, reverse proxies, dropped packets between tiers. CloudWatch and Dynatrace integration mean the observability your team already pays for shows up alongside the load test instead of in a separate tab.
.NET/IIS/ASP. About sixty Pro counters on top.
The .NET agent runs on the Pro tier and collects the base set of OS metrics alongside platform-specific counters from ASP.NET, the .NET CLR, IIS, and the combined IIS plus ASP.NET plus Application Pool process group. The dashboard surfaces them the same way it surfaces CPU, so a degradation in Lock Contention Queue Length plots against response time without you having to set anything up.
- ASP.NET (27 counters): Active Sessions, ASP.NET Application Restarts, Application Restarts, Cache Entries, Cache Hit Ratio, Cache Hits, Cache Misses, Cache Turnover Rate, Compilations, Current Requests, Disconnected Requests, Errors/sec, Last Request Execution Time, Pipeline Instance Count, Queued Requests, Rejected Requests, Request Wait Time, Requests Executing, Requests Not Found, Requests Not Authorized, Requests Queue Length, Requests Succeeded, Requests Timed Out, Requests/sec, Sessions Timed Out, Sessions Abandoned, Unhandled Execution Errors/sec.
- .NET CLR (15 counters): Class Loader Heap Size, Current AppDomains, Current Assemblies, Exceptions/sec, Heap Size, Generation 0 Collections, Generation 1 Collections, Generation 2 Collections, Lock Contention Queue Length, Lock Contention/sec, Lock Contentions, Logical Thread Count, Percent Time in GC, Physical Thread Count, Recognized Thread Count.
- IIS (15 counters): CGI Requests/sec, Connection Attempts/sec, Current Connections, Extension Requests/sec, File Cache Entries, File Cache Hits, File Cache Hit %, File Cache Misses, Kernel URI Cache Hit %, Kernel URI Cache Misses, Requests/sec, URI Cache Entries, URI Cache Hits, URI Cache Hit %, URI Cache Misses.
- IIS plus ASP.NET plus Application Pool process (5 counters): Handle Count, Private Bytes, Thread Count, Virtual Bytes, Percent Processor Time.
Server Performance. What is saturated across the fleet.
When a test runs against many hosts, the Server Performance overview lays them out side by side. Three top-of-page stat cards summarize peak CPU, memory, and network across all monitored servers. The chart plots whichever metric you pick across all of them in one view. For the test shown, peak CPU hits 97 percent, peak memory hits 85 percent, peak network hits 3.2 Gb/s. The table below names which host hit each peak and how many samples it contributed. Find the saturated host fast.
Individual Servers. Drilldown without mental math.
Click into any server and the view changes to a multi-axis chart of related metrics. The Summary group plots CPU, memory, and bandwidth on three axes against the same x-axis. The Processor group swaps in context switches and process queue length. Equivalent grouped views exist for memory, disk, network, and TCP. Every chart uses the same x-axis as the rest of the dashboard, so you can read server curves against response time without converting between time and user-level views in your head.
This is what feeds the AI bottleneck analysis.
The Analytics Dashboard runs correlation math on every monitored metric on every server against response-time degradation. Any resource whose curve correlates above 0.94 with the slowdown gets flagged. The AI report on the Report tab takes those flags and writes the bottleneck-source paragraph: which resource saturated, on which host, at which user level the response time went vertical. None of that synthesis works without the underlying monitoring on this page.
See the AI bottleneck analysis on the Analytics Dashboard page
The agent itself.
Multi-platform.
Agents for Windows, Linux, AIX, and .NET/IIS. Each agent collects the OS-native metrics for its platform and exposes them through the same vocabulary the dashboard already speaks. A .NET host and a Linux database server show up side by side in the Server Performance overview with no special configuration.
Fault tolerant.
Load tests have a way of breaking the things being tested. Servers crash, networks lose bandwidth, agents get disconnected. The agent writes every sample to local disk continuously, so if it loses its connection back to WPLoadTester (or the server itself goes down and comes back), the data survives. When the agent reconnects, WPLoadTester pulls the missing samples and merges them into the test results.
Firewall friendly.
Two modes. Live mode does secure traversal through a configurable port if your network team will open one. Offline mode runs the agent as a command-line tool: start it before the test, stop it when the test ends, copy the log files into WPLoadTester for post-test import. The offline path works through any firewall because it does not need a live connection in the first place.
Auto-detected.
Agents on your local LAN show up in WPLoadTester automatically. Remote agents add by IP address with one click. No manual server configuration, no agent registry to maintain. Upgrade an agent by clicking Update and the new version downloads itself.
Run a load test with monitored servers and the bottleneck names itself.
Server Monitoring ships in every WPLoadTester 7 install. Free captures CPU, memory, and disk; Pro captures the full 24-metric set plus AWS CloudWatch and Dynatrace integration. Request the beta to see your own server metrics correlated against response-time degradation in the Analytics Dashboard.
Comparing tiers? See the Free vs Pro split.