[Nottingham] Linux VM performance

Wed May 18 10:38:49 UTC 2016

Daryl, 
A lot of that went over my head, but have you tried using kvm instead of 
VirtualBox? I'm assuming you're using Linux as a host system, here. I've 
recently started messing about with kvm and it's renowned to have better 
performance, since the VM plugs into a kvm stub in the host kernel, which 
apparently reduces the amount of abstraction needed and runs close to 'bare 
metal' speeds. 
I also learned that you can bring up a terminal in the client machine by 
simply using 'virsh console....'. This was after messing about needlessly 
setting up ssh server! :-) 
Neal. 

--
Securely sent with Tutanota. Claim your encrypted mailbox today!
https://tutanota.com

18. May 2016 11:30 by nottingham at mailman.lug.org.uk:

> WARNING: Random meandering follows...
> I have a talk tomorrow evening. I have absolutely nothing to show yet.
> However, I have been having AWESOME (I spend too much time around an 
> American) fun tidying up and rejigging my performance analysis. I did a 
> talk on this last year.
> It's a simple scheduling algorithm so all tight loops and floating point 
> maths. It SHOULD suit JITs very well. The code size is small enough that it 
> will fit into CPU caches.
> The code is as identical as possible across languages. I'm generally 
> avoiding code that is too idiomatic; however, I'm using immutable 
> structures where possible and recursive techniques in those languages that 
> support it. I'm writing everything out to the terminal to stop those clever 
> lazy languages avoiding calculating anything (I'm looking at you Haskell!)
> Results below:
> Blitzmax: 0:24.87 real  23.99 user  0.24 sys
> Monkey-X: 0:15.84 real  15.17 user  0.24 sys
>  C# Mono: 1:01.38 real  58.84 user  0.83 sys
>      C++: 0:12.73 real  12.05 user  0.21 sys
>  Clojure: 1:01.01 real  62.52 user  1.01 sys
>        D: 0:32.22 real  31.98 user  0.15 sys
>  F# Mono: 0:58.05 real  57.32 user  0.52 sys
>       Go: 0:23.41 real  23.19 user  0.17 sys
>  Haskell: 0:38.96 real  38.27 user  0.50 sys
>     Java: 0:30.41 real  30.43 user  0.25 sys
>    OCaml: 0:43.50 real  43.19 user  0.20 sys
>  Python2: 4:46.03 real  282.67 user  1.43 sys
>  Python3: 4:26.41 real  264.80 user  0.72 sys
>    PyPy2: 0:16.76 real  16.11 user  0.20 sys
>    PyPy3: 0:32.44 real  31.34 user  0.34 sys
>     Ruby: 2:31.46 real  149.54 user  0.54 sys
>     Rust: 0:12.47 real  11.94 user  0.14 sys
>    Scala: 0:31.63 real  31.25 user  0.66 sys
> Ignoring the first two rows (which are proprietary game BASICs, but great 
> fun for chucking together graphical demos), then the most interesting 
> results are that Rust is quicker than Boost C++ and that PyPy is very 
> nearly as quick as both.
> I'm not comparing against .NET C# and F#. I would expect those to perform 
> much quicker, but of course they don't run on Mac OS/Unix/Linux! I would 
> imagine C# would be in the 20-25 second range.
> Anyway. that wasn't the point of my email. I was setting up a VirtualBox VM 
> to package all this up and so installed the latest Ubuntu (as I like to try 
> new shiny stuff). Performance was really poor. Ubuntu has always been a bit 
> graphics heavy and the acceleration isn't so good in VirtualBox it seems.
> So I'm going to do what I've done many times before and install a 
> non-graphical Debian. I can run Emacs through there for code editing and 
> Synaptic for package management. I can flick between virtual terminals with 
> Ctrl-Alt-Fn (IIRC) and configure it to run a decently high resolution using 
> GRUB.
> It sounds like an interesting experiment. How productive is it possible to 
> be using the terminal? In the modern world of HTML emails and 
> web-everything, it seems impossible to be productive (I'm not talking 
> sysadmin activity here). Email? Word processing using Latex? Spreadsheets 
> are fundamentally text based anyway so should be okay. Presentations would 
> be an obvious no-go I expect. What compromises need to be allowed? Full 
> screen image/PDF viewing perhaps.
> It might be an interesting experiment to undertake. Especially for someone 
> who can code up missing pieces as needed. Especially for an old-school guy 
> who considered the first GUIs to be a nice fad.
> "Someone" should take this on as an experiment and do a talk on it. I'm 
> thinking a three month experiment.
> Daryl.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.lug.org.uk/pipermail/nottingham/attachments/20160518/f2f6892a/attachment-0001.html>