[OpenWrt-Devel] SDK vs. Toolchain+

Felix Fietkau nbd at openwrt.org
Wed Jan 6 13:39:04 EST 2016

On 2016-01-06 17:52, Daniel Dickinson wrote:
> Hi,
> On 06/01/16 05:24 AM, Bastian Bittorf wrote:
>  > * Daniel Dickinson <openwrt at daniel.thecshore.com> [06.01.2016 11:11]:
>  >> Not sure where you got those numbers.  Did you also post the exact
>  >> hardware and software configuration, and testing methodology, and how
>  >> you did the timing and what numbers 140 is (real command to done
>  >
>  > check this thread:
>  > 
> https://lists.openwrt.org/pipermail/openwrt-devel/2015-December/037775.html
>  > 
> https://lists.openwrt.org/pipermail/openwrt-devel/2015-December/037856.html
> Obviously your idea of cheap, fast hardware is based on the assumption 
> that one is working on OpenWrt for pay rather than hobby. You're talking 
> about at least $1200-1600 CAD which is a lot of money these parts, for 
> the type of hardware you're talking about.
> When I did have a job that involved work on openwrt (which I'm not 
> entirely sure was worth it, for all that I got a couple gigs it wasn't 
> with open source-minded companies and for that and other reasons was 
> more frustrating than rewarding, for the most part, given that for 
> openwrt was something of personal interest and I found the proprietary 
> business interested tainted it for me, they say never mix business with 
> pleasure, and for good reason).  I did buy better hardware, but that is 
> now only decent, not the class of hardware you're talking about in your 
> email (and while good, wasn't real server class, ever). At the moment 
> OpenWrt is a hobby for me, so shelling out that kind of money, and 
> finding a place for such hardware, isn't an option.
> You're numbers are pretty bogus for the typical openwrt community 
> member, although they ought to be importance to those who can justify 
> that kind of investment as a business expense.
> So the work on improving build speed, based on Felix's suggestion is 
> still definitely something I think is worthwhile for the kind of 
> situation I am interested in.
So I went ahead and implemented that. Here are the improvements when
there are no kernel config changes (tested on ramips mt7621, where it
already uses the new image building code for everything):

target/compile: 12s -> 8s
target/install: 68s -> 16s

- Felix
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org

More information about the openwrt-devel mailing list