Early Ubuntu Hardware/Software Survey Data

[torsionbar28]

A bit odd they count how many CPUs you have, but not cores or threads.

I don't find it strange at all. Unlike games (e.g. Steam survey) that are optimized for high clocks and low core counts, a general purpose Linux OS will benefit from as many cores as you can throw at it. I can see where game devs care about core counts, since historically most games have not been written to take advantage of more than 4 cores (and many even less than 4). If 90% of the install base is using 4 cores, why spend the money to thread the game beyond 4 cores? For an OS vendor though, what difference does it make whether the user has 4 cores or 8 or 16? We all know the Linux OS will schedule processes accordingly.

What an OS vendor might care about however, is socket count. One socket tells us it's probably a consumer grade desktop or laptop peecee. Two sockets is a professional workstation or small/medium server. Four+ sockets is a big honkin mega server. Knowing this gives them valuable info on where their OS is being deployed.

[M@GOid]

Why so long? My Arch installs only take 10 and that's to get a full desktop up and running.

A look at the variables might explain:

- How is the size of the ISO file on Arch? Ubuntu is 1.8GB;

- What is the media being used? USB flash drive is faster than a DVD. A fast USB 3.0 driver will fly during a install;

- What CPU and drive are being used? A Atom with a HDD will be a lot slower than a i3/i5 with a SSD;

- Are you messing with the options of the installer? I myself never use the default options;

- Are the installer pulling packages from the internet? Ubuntu always pull packages from the internet in my installations.

I can install Ubuntu really fast, but the variables above have a say on the total time of any installation.

[starshipeleven]

A bit odd they count how many CPUs you have, but not cores or threads.

I'm betting it's because their data collection script is garbage. The examples of collected data from Michael back when he tested this were full of holes.

EDIT: maybe it's only their data mining script (the script that extracts data from the reports) that sucks.

[starshipeleven]

Not sure the meaning of Single core? does that mean these people are running laptops? or AMD cores/Intel no HT?

Single CPU usually (and in this case too) means single physical CPU, not single core CPU.

There are servers that can run 2 or 4 physical CPUs on the same motherboard (each with its own RAM and PCIe lanes, but working together under the same OS).

They say "We haven’t broken this down to cores but is something we will look in to."

[calc]

Why so long? My Arch installs only take 10 and that's to get a full desktop up and running.

Probably old systems, I just did a test install of 18.04 and from system power on to reboot took less than 6 minutes. Probably closer to 5 minutes in the installer itself.

[PackRat]

Where is the desktop environment statistics? I can't see gnome 3 being popular with people with 4 gb ram.

[olympus]

Where is the desktop environment statistics? I can't see gnome 3 being popular with people with 4 gb ram.

As far as I know the survey is only on Ubuntu not in Ubuntu flavours.

[PackRat]

As far as I know the survey is only on Ubuntu not in Ubuntu flavours.

The survey is in mate too.

[schmidtbag]

I don't find it strange at all. Unlike games (e.g. Steam survey) that are optimized for high clocks and low core counts, a general purpose Linux OS will benefit from as many cores as you can throw at it. I can see where game devs care about core counts, since historically most games have not been written to take advantage of more than 4 cores (and many even less than 4). If 90% of the install base is using 4 cores, why spend the money to thread the game beyond 4 cores? For an OS vendor though, what difference does it make whether the user has 4 cores or 8 or 16? We all know the Linux OS will schedule processes accordingly.

I mostly agree, but my point was more a matter of how the data is acquired in the first place. You can easily get number of cores/threads just from /proc/cpuinfo or lscpu (or even simpler, nproc, if the system has that command installed). I'm not aware of built-in functionality that can easily tell you how many CPU sockets you have.

In other words, you can pretty effortlessly figure out how many cores/threads a system has, but it's not so easy to identify how many CPUs you have, and yet Canonical didn't gather the easy-to-get info while going out of their way for the harder-to-get info.

What an OS vendor might care about however, is socket count. One socket tells us it's probably a consumer grade desktop or laptop peecee. Two sockets is a professional workstation or small/medium server. Four+ sockets is a big honkin mega server. Knowing this gives them valuable info on where their OS is being deployed.

Just being devil's advocate: you could have a dual-socket motherboard with only 8 total cores and 8 threads, meanwhile there are single-CPU systems with 32 cores and 64 threads. Sure, the dual-socket setup was more advanced for its time, but the 32-core system is obviously more capable. There is some validity in your point, but if Canonical doesn't have all the data, they're not getting the whole picture.

[olympus]

The survey is in mate too.

I didn't know that. Is it on all flavours? Only if it's in all flavours, desktop environments statitics would make sense.