#97872 apt: download and install in parallel

Package:
apt
Source:
apt
Description:
commandline package manager
Submitter:
David MarínCarreño
Date:
2026-11-07 21:29:02 UTC
Severity:
wishlist
#97872#5
Date:
2001-05-17 23:34:23 UTC
From:
To:
I've got a dial-up Internet connection, and, when I make apt-get upgrade,
I must wait some hours until any upgrade is done.

I think that it would be a good idea if apt were able to install packages
while downloading others. That is:


# apt-get upgrade
...
Downloaded package foo_1.0.i386.deb
Downloaded package libfoo_1.0.i386.deb
[Installing foo...][Downloading libc...]   [35%]
...

The package foo depends on libfoo. It downloads foo first, then libfoo,
and after this, it starts foo instalation while downloading another
package (such as libc, for example).

What do you think?

#97872#20
Date:
2026-11-07 21:27:16 UTC
From:
To:
Dear maintainers,

I would like to add a measurement to this old wishlist report, and ask
which part of it, if any, is still considered worth pursuing.

I wrote a small script that, for a given transaction, times the download of
all archives and then the sequential unpack with dpkg into a throw-away
root (no root privileges, nothing installed on the host). It does not
measure configure/maintainer scripts. One run on apt 3.3.3 / dpkg 1.23.11
(Kali rolling), installing libreoffice-writer with --no-install-recommends,
over an ordinary home connection:

packages: 51
size: 97 MB
download: 262 s
unpack: 7 s
total: 269 s
upper bound on the gain from overlapping download and unpack: 3 %

So on this link the download dominates and overlapping unpack with download
would save very little. This is a single run on one link; on a fast network
or with packages with long maintainer scripts the picture would differ, and
I can post more runs if useful.

What the numbers do suggest is that, for slow links, the time is spent in
the fetch phase. I would therefore like to ask:

1. Is there still interest in the original request (unpack while
downloading), given that the benefit depends heavily on the link speed?
2. Would better documentation and tuning of the per-host download behaviour
(Acquire::Queue-Mode, connections per host) be a more useful direction than
overlapping the two phases?

I am happy to send more measurements, or a patch for whichever direction
you consider acceptable, as a merge request on Salsa. If you think this
report should simply be closed, that is fine too.

Related reports: #55044.

Thanks for your time and for maintaining apt.

Enzo Abyie