linuxgeek: Re: [LINUX_Newbies] apt-fast or apt-get

Tuesday, April 21, 2015

Re: [LINUX_Newbies] apt-fast or apt-get

On Tue, Apr 21, 2015 at 12:37 PM, Gene Henley mhenley2@verizon.net
[LINUX_Newbies] <LINUX_Newbies@yahoogroups.com> wrote:
> What is your opinion of apt-fast instead of apt-get?

Never heard of it before... so I took a look... it's a wrapper around
aptitude or apt-get that uses aria2c to create concurrent downloads.

Keep in mind you now have a sample set of one, but my impression was
quite "meh".

I did an update and tried apt-fast configured for 10 and 5 concurrent
downloads and apt-get. In all three cases, I used -d to ensure a
download only so the same packages were downloaded each time. This is
on a 10Mb ADSL line.

With apt-fast configured to support 10 concurrent downloads:
sudo time apt-fast -d dist-upgrade
42.86 user
15.33 system
13:34.42 elapsed
7% CPU (0 avg text + 0 avg data 58600 max resident)k
180256 inputs + 1443240 outputs
(475 major + 357908 minor) pagefaults
0 swaps

With apt-fast configured to support 5 concurrent downloads (the default):
41.18 user
16.13 system
13:47.50 elapsed
6% CPU (0 avg text + 0 avg data 58600 max resident)k
562144 inputs + 1435744 outputs
(954 major + 357607 minor) pagefaults
0 swaps

Standard apt-get
sudo time apt-get -d dist-upgrade
73.50 user
32.72 system
13:37.86 elapsed
12% CPU (0 avg text + 0 avg data 58764 max resident)k
11704 inputs + 1435408 outputs
(39 major + 21938 minor) pagefaults
0 swaps

Note that they all ran for pretty much the same amount of elapsed
time, with apt-get actually coming in faster than apt-fast at 5
concurrent threads and just three seconds slower than apt-fast at 10
threads.
Additionally, the amount of pagefaults for both apt-fast runs greatly
outnumber those for standard apt-get at one package at a time.

CPU Usage was also very similar across them with apt-get being
slightly higher, but not outrageously so.

So from my very limited testing, I saw no benefit really to using
apt-fast over apt-get. Remember that your download ability is only as
good as your connection. So 10Mb for one file vs 1Mb for 10 files,
the bandwidth is still the same. This may be more noticeably faster
if you're on FIOS or something that's truly fast... I don't have
access to a fibre connection to run this on right now. Hmmmm maybe
I'll have to try that in digital ocean...

So I did launch a DO droplet with arguably gigabit speed to the local
repository mirrors. The Droplet was 14.04 and showed the following
updates:
107 upgraded, 4 newly installed, 0 to remove and 0 not upgraded

apt-fast configured for 10 concurrent downloads:
2.92 user
1.04 system
0:08.13 elapsed
48% CPU (0 avg text + 0 avg data 54456 max resident)k
5544 inputs + 178792 outputs
(22 major + 146884 minor) pagefaults
0 swaps

apt-fast configured for 5 concurrent downloads:
3.13 user
1.08 system
0:06.04 elapsed
69% CPU (0 avg text + 0 avg data 54436 max resident)k
0 inputs + 178800 outputs
(0 major + 144554 minor) pagefaults
0 swaps

Standard apt-get:
4.41 user
0.50 system
0:11.08 elapsed
44% CPU (0 avg text + 0 avg data 54560 max resident)k
0 inputs + 178744 outputs
(0 major + 18068 minor) pagefaults
0 swaps

So on the DO Droplet, with very fast networking, apt-get came out
slowest at 11 seconds, but that's really not THAT much faster than
apt-fast's fastest time of 6 seconds for 5 threads or 8 for 10
threads. It actually IS 50% faster roughly at 5 threads, but 5
seconds difference in reality is still not really that much.

And again, we still see the crazy number of pagefaults compared to
apt-get, CPU usage is more in this scenario, but still relatively even
for 10 threads vs apt-get, and way higher for 5 threads of apt-fast.

Interesting... and completely unscientific, YMMV... it's neat but my
results don't look quite so earth shattering. I expected more when I
read the github info and it said apt-fast can "drastically improve apt
download times by downloading packages in parallel, with multiple
connections per package."

With tweaking, it's likely you can get better results, for sure, but
this is a closer apples to apples comparison, IMO, I used no tweaking
at all, save trying the default number of concurrent threads and then
10 for apt-fast.

Cheers
Jeff