aria2
Big files, slow servers, many downloads, torrents
- Up to 10× faster when a server limits each connection
- BitTorrent, Metalink and mirrors built in
- A download server you control over RPC
Measured comparison
aria2 vs wget vs curl: measured, not guessed
All three download files from the command line, but they are built for different jobs. We ran them side by side on the same machine and the same test servers, and wrote down every number.
The short answer
wget
Simple downloads and copying whole websites
- Smallest memory use in our test
- Recursive mirroring with -r
- Preinstalled almost everywhere
curl
APIs, uploads and scripting any protocol
- 25 protocols in the build we tested
- Uploads, forms and full HTTP control
- Remember -L for redirects
One big file from a server that limits each connection
Mirrors and file hosts often cap the speed of a single connection. Our test server allowed 2 MiB/s per connection on a 24 MiB/s line. File size: 50 MB.
Show as a table
| Tool | Command | Seconds |
|---|---|---|
| wget | 24.0 | wget URL · 24.0 s |
| curl | 24.0 | curl -O URL · 24.0 s |
| aria2, 1 connection | 24.1 | aria2c URL · 24.1 s |
| aria2 -x16 -s16 | 12.2 | aria2c -x16 -s16 URL · 12.2 s |
| aria2 -x16 -s16 -k1M | 2.3 | aria2c -x16 -s16 -k1M URL · 2.3 s |
With one connection all three take 24 seconds; the server decides. aria2 with 16 connections took 12.2 seconds, because its default -k 20M splits a 50 MB file into only 2 pieces. With -k1M it used all 16 connections and finished in 2.3 seconds.
20 files at once
Twenty 2 MB files from a server that adds 80 ms of latency per request and caps each connection at 2 MiB/s. wget and plain curl fetch one file after another; curl --parallel and aria2 fetch several at a time.
Show as a table
| Tool | Command | Seconds |
|---|---|---|
| wget -i | 21.8 | wget -i urls.txt · 21.8 s |
| curl, one by one | 21.9 | xargs -n1 curl -O < urls.txt · 21.9 s |
| curl --parallel | 5.5 | curl --parallel --parallel-max 5 -O … · 5.5 s |
| aria2c -i | 4.5 | aria2c -i urls.txt · 4.5 s |
| aria2c -i -j16 -x4 | 2.0 | aria2c -i urls.txt -j16 -x4 -k1M · 2.0 s |
aria2 runs 5 downloads at a time by default, a bit faster than curl --parallel with the same limit. With -j16 -x4 it reached the line limit: 2.0 seconds against 21.8 for wget. All five runs produced identical files.
When the server does not limit connections
On a local server with no limit, all three downloaded the 50 MB file in under a second, and wget and curl were a little quicker than aria2, which first reserves disk space for the file. If one connection already fills your line, extra connections cannot add speed.
Memory
Measured during the same HTTP download. "Own memory" is what the program allocated for itself; the total adds shared system libraries.
ToolOwn memoryTotal (RSS)
wget
curl
aria2
wget is the leanest. aria2 needs more because it keeps a download engine, a disk cache and the BitTorrent code ready, but 4.5 MiB is still small: the official aria2 figures are 4–9 MiB.
Resuming an interrupted download
We stopped each tool halfway through the throttled download and started it again. All three finished with a correct file: wget with -c, curl with -C -, aria2 with -c. aria2 even resumed without -c, because its .aria2 control file records which pieces are done. Note that aria2 reserves the full file size first, so a partial file already shows its final size.
Redirects: the curl trap
Many download links redirect to a mirror. wget and aria2 follow redirects by default. curl does not: without -L it saved an empty file and still exited with code 0, so a script would not notice. Always use curl -L for downloads.
What each tool can do
Checked against the documentation of each tool and, where possible, by running it.
| aria2 | wget | curl | |
|---|---|---|---|
| One file over many connections | Yes | No | No |
| One file from several mirrors at once | Yes | No | No |
| BitTorrent and magnet links | Yes | No | No |
| Metalink | Yes | Some builds | No |
| Resume an interrupted download | Yes | Yes | Yes |
| Remote control (RPC) | Yes | No | No |
| Mirror a whole website | No | Yes | No |
| Upload, 25+ protocols, APIs | No | No | Yes |
| Preinstalled on most Linux systems | No | Yes | Yes |
The same task in all three
We ran every row except the last two against our test server, with our own values for the numbers; each produced the correct file.
| Task | wget | curl | aria2c |
|---|---|---|---|
| Download a file | wget URL | curl -O URL | aria2c URL |
| Save under another name | wget -O name.iso URL | curl -o name.iso URL | aria2c -o name.iso URL |
| Save into a folder | wget -P dir URL | curl --output-dir dir -O URL | aria2c -d dir URL |
| Resume | wget -c URL | curl -C - -O URL | aria2c -c URL |
| Limit the speed to 1 MiB/s | wget --limit-rate=1m URL | curl --limit-rate 1M -O URL | aria2c --max-download-limit=1M URL |
| Retry up to 10 times | wget --tries=10 URL | curl --retry 10 -O URL | aria2c --max-tries=10 URL |
| Send a header | wget --header="X-Key: 1" URL | curl -H "X-Key: 1" -O URL | aria2c --header="X-Key: 1" URL |
| Set the user agent | wget -U "Agent/1" URL | curl -A "Agent/1" -O URL | aria2c -U "Agent/1" URL |
| HTTP login | wget --user=U --password=P URL | curl -u U:P -O URL | aria2c --http-user=U --http-passwd=P URL |
| Download a list of URLs | wget -i urls.txt | xargs -n1 curl -O < urls.txt | aria2c -i urls.txt |
| Skip the certificate check | wget --no-check-certificate URL | curl -k -O URL | aria2c --check-certificate=false URL |
| Follow redirects | wget URL | curl -L -O URL | aria2c URL |
| One file over 16 connections | Not possible | Not possible | aria2c -x16 -s16 -k1M URL |
| Download a torrent | Not possible | Not possible | aria2c file.torrent |
| Copy a website | wget -r -np URL | Not possible | Not possible |
Which one should you use?
- A large file from a slow or busy server
aria2c -x16 -s16 -k1M - Hundreds of files from a list
aria2c -i urls.txt -j16 - A torrent, magnet link or Metalink
aria2c - A download queue you control from a browser or phone
aria2c --enable-rpc with a WebUI - A whole website or directory tree
wget -r -np - An API call, an upload or a form
curl - A quick one-off download on a server you just logged into
wget or curl -L, already installed
Questions
Is aria2 always faster than wget?
No. aria2 is faster when the server limits each connection or when many files download at once: up to 10× in our tests. When one connection already fills your line, wget and curl are just as fast, and were slightly quicker on our unlimited local server.
Can curl download one file with several connections?
Not by itself. curl --parallel downloads several URLs at the same time, but each file still uses one connection. aria2 splits one file into pieces and fetches them in parallel.
Can I use aria2 instead of wget in scripts?
For downloads, yes: aria2c -o name -d dir URL covers what wget -O and -P do, and the exit code tells the script what went wrong. aria2 cannot copy whole websites; keep wget for that.
Which uses less memory?
wget: 0.7 MiB of its own memory in our test, curl 1.5 MiB, aria2 4.5 MiB. All three are small enough for a Raspberry Pi or a router.
Does wget support BitTorrent?
No. Neither wget nor curl downloads torrents or magnet links. aria2 is a full BitTorrent client as well as an HTTP and FTP downloader.