Anchor Download Manager: The ONLY real IDM alternative!

Creative Geek•

There are thousand of download managers out there, none of them buys you the guarantee of a completed download, see what it takes to build an Unbreakable Download Manager.

Anchor Download Manager: The ONLY real IDM alternative!

If you have ever downloaded a 20 GB archive on a metered, flaky internet connection, you know this specific heartbreak:

The progress bar creeps to 97%. You have spent four hours babying the connection, burning a chunk of your monthly quota. Then the connection drops for two seconds. You click "Resume."

The browser spits out "Failed — Network Error", resets the counter to 0 MB, and expects you to start all over again.

Here is the brutal truth that most software engineers in fiber-connected tech hubs forget: 97% is not 97%. If a download fails at 97%, it is 0%.

In regions where internet is strictly metered—like Egypt, where households budget around a fixed ~300 GB monthly quota—a failed download isn't a minor annoyance. It is a direct financial loss and a wasted quota slice that you cannot get back.

When I started building Anchor Download Manager (ADM), the objective was not to chase vanity download speeds or build an unreadable matrix of buttons. The objective was simple: a download you start must finish, no matter how hostile the network or the server is.

Here is a deep dive into the real mechanics of download managers, the edge case from hell that ruins multi-connection downloads, and how we solved it.

The Anatomy of Speed: Connection Ramping

Why do browsers crawl at 2 MB/s on a link that should deliver 20 MB/s?

Web browsers typically fetch files over a single TCP stream. If the remote server throttles per-connection bandwidth, or if packet loss triggers TCP’s congestion avoidance window to collapse, your single stream slows down to a crawl.

Accelerated download managers solve this using HTTP Range headers, splitting a file into segments and downloading them simultaneously.

Anchor Download Manager Main Window

In ADM, we don't just blast a server with 16 connections right away (which often gets your IP temporarily rate-limited). We use connection ramping:

  1. We establish Connection 1 and probe the server (does it support Accept-Ranges: bytes? What is the Content-Length?).
  2. Once the server confirms range support, ADM dynamically ramps connections: 1 → 2 → 4 → 8.
  3. As high-speed connections complete their allocated segments, remaining segments are dynamically split so idle connections take over unread portions of the file.

alt

The Edge Case From Hell: Mid-Flight File Mutation

Imagine this real-world scenario:

You are downloading an 8 GB game patch or an OS image from a busy server or CDN. The download takes 30 minutes.

While your download is at 45%, the server admin deploys a hotfix or updates the file at the exact same URL. Or, your traffic routes to a different CDN edge node that is running one revision ahead.

Now look at what happens inside a naive multi-connection manager:

autonumber. participant Engine as Download Manager. participant C1 as Conn 1 Running. participant C2 as Conn 2 Spawning. participant Srv as Server / CDN. Note over Engine, Srv: Server initially serves Revision A. Engine- C1: GET /file.iso Range: 0 - 2GB. C1 Engine: Streaming Revision A bytes.... Note over Srv: ⚠️ File is silently updated on server to Revision B!. Note over Engine: Conn 1 finishes chunk — needs more work. Engine- C2: Split remaining chunk! GET /file.iso Range: 2GB - 4GB. C2 Engine: Streaming Revision B bytes!. Note over Engine: 💥 Disaster: Stitches Revision A + Revision B into one file

Connection 1 finishes its first chunk. It wants to maximize throughput, so it splits the next remaining chunk and spawns a new HTTP request for bytes 2,147,483,648 onwards.

If the manager naively issues a fresh Range request without verification, Connection 2 begins downloading bytes from the new file (Revision B), while your disk already contains the first half from the old file (Revision A).

At 100%, the download manager happily announces "Complete!" But the archive is corrupt. The checksum fails. You just burned 8 GB of data to produce digital garbage.

How ADM Prevents This: Overlapping Seam Verification

To prevent this silent corruption without penalizing download speeds, ADM implements Overlapping Seam Verification:

Start "Conn 1 finishes chunk & seeks split" Split "Calculate split boundary on target chunk". Split Probe "Spawn next connection with slight byte overlap". Probe Wire "Read first 64 bytes off the wire at seam". Wire Compare "Do seam bytes match\nexisting downloaded bytes?". Compare -- "Yes File unchanged " Parallel "Commit bytes & continue multi-stream ramp". Compare -- "No File mutated on server! " Collapse "Trigger Single-Stream Collapse". Collapse StopNew "Halt spawning any new connections". Collapse Drain "Let active child connections finish their assigned blocks". Collapse SoleSurvivor "Original Conn 1 continues streaming remainder of entire file". SoleSurvivor Done "File completes cleanly as 100% coherent original revision"

Here is how it works under the hood:

  1. The Overlap Sample: When a connection takes over a seam, ADM doesn't blindly trust the new response. It samples a small buffer of bytes (up to 64 bytes) overlapping the known boundary.
  2. Non-blocking verification: This verification happens in memory while the parent connection is held active.
  3. The Single-Stream Collapse: If the bytes at the seam do not match, ADM knows the file underneath has mutated. Instead of destroying the download or throwing up an error dialog:
    • It immediately stops creating new connections.
    • It allows active connections to safely finish writing the chunks they already pulled.
    • Crucially, Connection 1’s original HTTP stream is kept alive. Because Connection 1's socket was opened against the original file revision, ADM routes the remainder of the download through that single surviving stream.

The result? You don't restart from zero. You don't burn extra bandwidth. And you don't end up with a Franken-file stitched together from two conflicting releases. You receive the exact, untorn file you started downloading.

The "Temp Folder" Trap: Direct-to-Destination Writes

If you have used legacy download managers on Windows, you have almost certainly encountered this absurd sequence:

  1. You set your download folder to an external hard drive (D:\Downloads).
  2. You download a 40 GB video file.
  3. The download reaches 99%... and your entire computer freezes.
  4. An error dialog pops up: Error: Drive C: is out of disk space.

Why did C: run out of space when you were downloading to D:?

Because traditional tools download temporary parts to %TEMP% on your primary drive (C:\Users\<Name>\AppData\Local\Temp\). When the download reaches 100%, they assemble the parts by reading from C: and writing to D:.

subgraph OldWay "Traditional Download Tools". direction TB. O_Streams "Parallel Streams" Write part files O_Temp "C:/AppData/Local/Temp/". O_Temp 99% Reassembly: Read & Copy across disks O_Dest "D:/Downloads/large_file.iso". O_Crash "❌ C: runs out of space br/ ❌ Double SSD wear br/ ❌ Agonizing merge delay" O_Temp. end. subgraph ADMWay "Anchor Download Manager". direction TB. A_Streams "Parallel Streams" Direct offset writes A_Anchor "D:/Downloads/large_file.iso.anchor". A_Anchor "Instant atomic rename 0 ms " A_Final "D:/Downloads/large_file.iso". A_Win "✅ Zero C: drive usage br/ ✅ Single SSD write pass br/ ✅ Zero reassembly time" A_Final. end

This design has three major flaws:

  • Catastrophic Failure: If your C: drive has less free space than the file size, the download dies at 99%.
  • Double SSD Wear: Every single gigabyte is written to your drive twice.
  • The Reassembly Hang: For huge files, reassembling segments can take 5 to 15 minutes after the network transfer has already finished.

ADM's Approach: Sparse Direct Allocation

ADM writes directly to the final destination.

When you start a download on D:\, ADM allocates a .anchor part file directly on D:\. Each connection writes its byte ranges into its exact physical offset in that single file.

When the last connection delivers its final byte, ADM flushes buffers (fsync), verifies the range map, and performs an instant atomic rename from .anchor to the final filename.

There is no temp folder. There is no second write pass. The moment network traffic hits 100%, your file is ready to open immediately.

alt

Surviving the Chaos Gauntlet: 10 Hostile Server Scenarios

Writing a download manager that works on a high-speed CDN with clean HTTP/1.1 compliance is easy. Building one that survives the real internet requires testing against broken, misconfigured, and hostile servers.

During development, we built a dedicated testing harness (crates/anchor-testkit) running a chaos HTTP server designed to torture the engine with every pathology observed in the wild:

  1. Single-connection hard limits: Servers that drop the second request with 403 Forbidden or TCP RST if you attempt multi-streaming. ADM falls back to a clean single stream without dropping bytes.
  2. Lying Content-Range: Servers that claim 206 Partial Content but send data from byte 0 anyway.
  3. Mid-stream stall & freeze: Connections that keep TCP sockets alive but stop transmitting bytes indefinitely.
  4. Sudden truncation: Servers that close connections prematurely without sending a FIN flag or signaling an error.
  5. Moving targets (Redirect chains): Mid-download authentication tokens expiring and redirecting to login walls.
  6. Corrupt range boundaries: Servers that return off-by-one byte ranges.
  7. Burst-throttled sockets: Servers that send 10 MB in a flash and then sleep for 10 seconds.
  8. Unannounced non-resumable endpoints: Handling dynamic downloads that silently drop resume support mid-way.
  9. Stale CDN caches: Edge servers delivering conflicting file sizes across parallel connections.
  10. File mutation at the seam: The exact scenario we detailed above.

Every single scenario was turned into an automated regression test in Rust. If an edge case can happen on a flaky connection in Cairo or a spotty train Wi-Fi in Europe, the engine must handle it autonomously without requiring the user to start over.

Reliability First, Everything Else Second

Anchor Download Manager is written in Rust with a native Qt 6 interface. It uses minimal system resources, avoids web-view bloat, and integrates directly with Chromium-based browsers (Chrome, Brave, Edge) via native messaging.

Downloads shouldn't be a gamble. You shouldn't have to monitor a progress bar wondering if your connection dropping for three seconds will erase two hours of waiting.

If you care about downloads that finish every single time:

To try the latest alpha release for Windows 10/11 x64, fill out this 6 MCQs form, you will be redirected to the download instantly: https://forms.gle/dULjuy8tVKZiSkLK9

alt