[syslinux] Confusion on lpxelinux vs. gpxelinux vs. ipxe vs gpxe.

[Doug Scoular]

Hi Michael,
Thanks for answering the substance of my questions. It looks like I should
be focussing on lpxelinux and taking a good look at iPXE itself.

Cheers,

Doug

On 25 October 2015 at 20:27, Michael Brown <mcb30 at ipxe.org> wrote:

> On 25/10/15 01:04, Gene Cumm wrote:
>
>> On Sat, Oct 24, 2015 at 6:15 PM, Michael Brown via Syslinux
>> <syslinux at zytor.com> wrote:
>>
>> Also, not a fork: http://git.ipxe.org/ipxe.git/commitdiff/8406115
>>>
>>
>> A fork is a fork, regardless the reasons behind it (yes, I have some
>> understanding in this case).  iPXE is based off of forking further
>> development as of a certain gPXE commit with some backporting of gPXE
>> development to iPXE.
>>
>
> From my perspective, a "fork" is when you take someone else's code and
> spin it off into a new project.  Taking what is primarily your own code and
> renaming it does not, for me, count as a "fork".  Maybe we're just using
> different definitions of the word.
>
> Anyway, to answer the questions of substance in this thread:
>
> There were some issues relating to iPXE real-mode operations that show up
> on pre-Westmere CPUs with newer (and stricter) versions of KVM.  This was
> fixed in a series of commits last year:
>
>   http://git.ipxe.org/ipxe.git/commitdiff/6d4deee
>   http://git.ipxe.org/ipxe.git/commitdiff/5a08b63
>   http://git.ipxe.org/ipxe.git/commitdiff/c64747d
>   http://git.ipxe.org/ipxe.git/commitdiff/bcfaf11
>
> along with some patches to avoid unnecessary real-mode transitions
> entirely (e.g. by allowing the system timer tick interrupt to happen in
> protected mode).
>
> A quick test just now under current KVM (running on a five-year-old Core
> i5) shows a download speed of around 1600Mbps sustained for 2.5GB of data
> downloaded from the VM host to the guest:
>
>   http://ipxe.org/_media/screenshots/ipxe_speed.png
>
>
> All of my GigE test hardware sustains 1000Mbps HTTP downloads under iPXE.
> My 10GigE test hardware typically sustains somewhere in the region of
> 2000-3000Mbps (CPU-limited, on a very old Core2Duo CPU).  Any new iPXE
> driver developed commercially will be tested and expected to reach 1000Mbps
> or have good reasons for not doing so.
>
>
> You can construct an ipxelinux.0 as an updated replacement for
> gpxelinux.0, to solve the problems relating to the old gPXE code present in
> gpxelinux.0.  However, there's not much motivation to do so. Instead, you
> should either use lpxelinux.0, or use iPXE itself.
>
> Michael
>



-- 

The big print giveth and the small print taketh away.