How do I find where time is spent in a page/API call?

Break the request into DNS, TLS handshake, time-to-first-byte (TTFB), and content download. Compare these steps across hosts and runs to see whether the delay is name resolution, negotiation, server think-time, or transfer.

What is BDP and why does it matter?

The bandwidth-delay product (BDP) is the amount of data needed in flight to fully utilize a path. If TCP windows or congestion control never reach the BDP, throughput will be capped. We estimate path BDP from RTT and observed rates to spot under-utilization.

Can you analyze HTTPS performance without decryption?

Yes. Even without payloads, TLS metadata and timing reveal resumption/0-RTT usage, ALPN results (h2/h3), handshake retries, and TTFB. With keys or pre-decrypted captures, we add header-level insights like cache/compression and response codes.

When should I use HTTP/2 or HTTP/3?

HTTP/2 multiplexing helps when many small objects are fetched over one path, but is sensitive to loss. HTTP/3 over QUIC avoids TCP head-of-line blocking and can improve performance on lossy or mobile links, unless UDP/443 is blocked causing fallback.

How do I run a fair performance test?

Use controlled runs (same client, same path, same object set), capture near the client and egress, include DNS and TLS, and compare baseline versus candidate changes. Repeat at peak and off-peak to avoid misleading results.

Performance Analysis | My Network Consultant

Overview

“It works” isn’t the same as “it feels fast.” Upload a PCAP/PCAPNG and let the analyzer decompose every request into DNS, TLS, TTFB, and download, then overlay transport efficiency (BDP usage, window scaling) and protocol features (TLS resumption/0-RTT, HTTP/2 prioritization, HTTP/3 retries/fallbacks). You’ll get clear guidance on what to fix first.

What you’ll see: timing breakdowns, redirect chains, cache/compression hints, header/cookie bloat, BDP utilization, H2 multiplexing quality, H3 fallback clues
Best for: speeding up pages/APIs, validating CDNs/migrations, catching regressions
Works with: origin & CDN edges, mobile/WAN paths, TLS 1.2/1.3, HTTP/2–3

What to look for

DNS & TLS: slow resolvers, handshake retries, missing resumption/0-RTT, ALPN downgrades
TTFB & think-time: back-end delay vs network; watch spikes at p95/p99
Transfer efficiency: BDP underuse, small receive windows, loss → poor throughput
Payload efficiency: no compression on text assets, missing cache headers, redirect chains
HTTP/2 behavior: too many sequential tiny streams → prioritization issues
HTTP/3 signals: QUIC handshake retries; UDP/443 blocked → fallback to H2/H1
Header/cookie bloat: oversized request headers driving latency and cache misses

Capture tips

Use controlled runs: same client, same object set, stable network.
Capture near the client (UX view) and, if possible, at egress (path effects).
Include DNS and TLS flows to correlate name and handshake timings with TTFB.
Disable NIC offloads (LRO/GRO/TSO/checksum) to preserve accurate timing and sizes.
Keep captures focused to fit the 10 MB free demo limit.

Example walkthrough

Filter to the app and host: e.g., (tcp port 443 or udp port 443) and host api.example.com plus DNS.
Review DNS → TLS → TTFB timing; confirm ALPN (h2/h3) and resumption.
Inspect redirect chains, status codes (if visible), and cache/compression hints.
Check BDP vs throughput (from RTT and observed rates) and window scaling.
Compare “before vs after” changes (even single capture): call out reduced handshake/TTFB or improved transfer rate.
Export a brief summary with the top two fixes and expected impact.

Targets & quick wins

DNS median < 100 ms; avoid TCP fallback except for very large answers.
TLS handshake < 300 ms on healthy paths; enable resumption/0-RTT where safe.
TTFB < 200–300 ms for cached content; minimize server think-time on APIs.
Compression: gzip/br on text assets; avoid double-compression on already-compressed media.
H2/H3: keep streams concurrent; fix prioritization for many small assets; ensure UDP/443 not blocked.

Performance Analysis FAQ

Break down each request into DNS, TLS, TTFB, and download; compare across hosts and runs to locate the bottleneck.

The bandwidth-delay product shows how much data must be in flight to fill the pipe; if windows never reach it, throughput is limited.

Yes—metadata and timing reveal resumption/0-RTT, ALPN, retries, and TTFB; decrypted captures add header-level insights.

H2 multiplexing helps with many small assets but is sensitive to loss; H3 can shine on lossy/mobile paths if UDP/443 is allowed.

Use controlled runs, include DNS/TLS, capture near the client and egress, and compare baseline vs change at peak and off-peak.