[Sugar-devel] Vision

[James Cameron]

On Wed, Apr 20, 2016 at 06:54:58PM -0400, Dave Crossland wrote:
> 
> Hi
> 
> On 20 April 2016 at 18:27, James Cameron <[1]quozl at laptop.org> wrote:
> 
>     On Wed, Apr 20, 2016 at 05:22:21PM -0400, Dave Crossland wrote:
>     >
>     > On 20 April 2016 at 16:46, James Cameron <[1][2]quozl at laptop.org> wrote:
>     >
>     >     the performance ratio between our low-cost
>     >     low-power hardware and the competition was already evident on Fedora
>     >     Linux; it didn't need Windows to expose it
>     >
>     > Sorry if this is an obvious question, but, can anything done to make
>     > Sugar feel faster on XO-1s today? 
> 
>     Yes, and I've been doing some of that in the past few months. 
> 
> AWESOME
>  
> 
>     With 13.2.7 you have my latest work, which added swap and removed several
>     animations.
> 
> Great :)
>  
> 
>     Adding swap has mostly removed memory pressure. 
> 
> Is it possible to configure the swap to run on the SD card, which, if failing
> due to thrashing, can easily be replaced?

When OLPC OS on XO-1 is installed on SD card, the swap partition is on
SD card automatically.  Because /var/swap is created from /dev/zero
during olpc-os-builder, is compressed to nearly nothing by the image
generation, and then expanded during install.  Be assured that the
string of zeros comes from the build host and is well travelled by the
time you get to overwrite it during page outswap.

>  
> 
>     Under memory
>     pressure, activity startup is roughly doubled,
> 
> WOW
>  
> 
>     as the CPU spends time thrashing in the memory management.  
> 
>     Disadvantage is higher power cost
>     and possibly decreased Flash endurance, although the endurance of a
>     set of heavily used XO-1 has shown no sign of the deterioration
>     expected by now.
> 
> Yes, it is somewhat astonishing to me that any XO-1 are still working at all :)
> I would have thought they'd all have cracked screens and ripped keyboards by
> now.
> 
> Maybe this was brought up on the XO-1 thread I started, but I didn't remember
> if so; does anyone has any suggestions about how many XO-1s are still in use?
>  
> 
>     Removing animations has allowed CPU cycles to be better spent on
>     responsiveness.  At one stage we had 50/50 competition between the
>     activity launch animation and the starting activity. 
> 
> Woah :)
>  
> 
>     Instrumenting
>     the frame and transition box animations showed there was enough time
>     for only one or two intermediate animation states before the final
>     state; which turned out to the cost of handling the function key
>     release event.  Some of these changes are not in Sugar master yet, but
>     in an OLPC branch; one such was proposed, but immediately closed with
>     appeal to process;
> 
>             [3]https://github.com/sugarlabs/sugar/pull/619
> 
> I sympathise with Sam's request to discuss further; eg, perhaps there is a
> compromise by wrapping the decision to animate in an if/else block that checks
> some cpuinfo in /proc, or if the number of running activities can be obtained
> very cheaply through a len() call or something?

Perhaps.  A fork is cheaper though.

>     As for what to do next; ideas are welcome, but here's a few;
> 
>     - profiling, of startup, of interactive response, (i've used xdotool
>       for interactive response tests),
> 
> AWESOME! Could you make a screencast showing how to set this up? 

Was documented in some bugs and pull request discussions.

>  
> 
>     - upgrade Gtk3, and GObject, to fix the memory leaks,
> 
> I like it!
>  
> 
>     - record metrics of response, deidentify, aggregate, and report.
> 
> I think this kind of data driven development is crucially important :) 
>  
> 
>     Although at this stage the interest in XO-1 should have degraded as
>     the units have degraded, and any return on investment is doubtful.
> 
> I'm not sure; even if the XO-1 units themselves are gone, the cheapest
> computers will always be puny, either (non-)refurbished clunkers passed down to
> kids, or $5 computers like the Raspbery Pi Zero. 
> 
> By optimizing for XO-1, we optimize for spending power of small
> children. 

A phone will be cheaper to buy, own, and run.

>     Plenty of people left who whinge about XO-1, but ask them to test a
>     patch or release and no response.
> 
> Kindly, this is because you assume too much technical skills/experience on
> their part, and I suspect that there is a "tact filter mismatch" per [4]
> www.mit.edu/~jcb/tact.html :) 

No, I disagree.  If they truly needed what they were whinging for,
they would figure out how to be more tactful.  e.g. by asking how to
test.

> This was/is also a brake on the speed of development of the fontforge
> community, which is written in C and has its own X toolkit; the community of C
> developers were generally unwilling to 'baby step' enthusiastic but
> inexperienced community peers in how to apply and test patches. 
>  
> 
>     So it's more about people wanting their rainbow pooing unicorns.
>     Unrealistic expectations, polarised framing, denial, and consequent
>     unwillingness to be involved.
> 
> I am again reminded of [5]https://youtu.be/N9c7_8Gp7gI?t=9m1s 9m1s to 10m23s as
> an amusing anecdote about how to work productively with people who are setting
> out to work against you :)
> 
> --
> Cheers
> Dave
> 
> References:
> 
> [1] mailto:quozl at laptop.org
> [2] mailto:quozl at laptop.org
> [3] https://github.com/sugarlabs/sugar/pull/619
> [4] http://www.mit.edu/~jcb/tact.html
> [5] https://youtu.be/N9c7_8Gp7gI?t=9m1s

> _______________________________________________
> Sugar-devel mailing list
> Sugar-devel at lists.sugarlabs.org
> http://lists.sugarlabs.org/listinfo/sugar-devel


-- 
James Cameron
http://quozl.netrek.org/