Some what In-Reply-To: <1323092879.2629413.1445740496536.JavaMail.yahoo at mail.yahoo.com> On Sun, Oct 25, 2015 at 02:34:56AM +0000, Patrick Masotta via Syslinux wrote:> Micheal Brown > > > > iPXE's HTTP implementation is definitely not slow. On a Gigabit LAN, > > you should get the full 1000Mbps speed for HTTP downloads: a 200MB test > > file should transfer in under two seconds. For most practical OS boots, > > you won't even have time to read the text that iPXE displays on-screen. > > > HTTP at line speed (1Gb) on an iPXE client? > I was never able to get anything even close to that.Tell about your own setup. Compare it with the other setup. See where it differs. Narrow the gap. Groeten Geert Stappers P.S. Reread https://en.wikipedia.org/wiki/Robustness_principle
>>>On Sun, Oct 25, 2015 at 02:34:56AM +0000, Patrick Masotta via Syslinux wrote:> Micheal Brown > > > > iPXE's HTTP implementation is definitely not slow. On a Gigabit LAN, > > you should get the full 1000Mbps speed for HTTP downloads: a 200MB test > > file should transfer in under two seconds. For most practical OS boots, > > you won't even have time to read the text that iPXE displays on-screen. > > > HTTP at line speed (1Gb) on an iPXE client? > I was never able to get anything even close to that.Tell about your own setup. Compare it with the other setup. See where it differs. Narrow the gap. Groeten Geert Stappers <<< Sorry Geert but I think the one that has to talk about his setup and the gear used in his measurements (Spirent?, Ixia?, a school ruler?) should be the one that claims the above mentioned fantastic figures. Best,Patrick
On 25/10/15 11:42, Patrick Masotta via Syslinux wrote:> Sorry Geert but I think the one that has to talk about his setup and the gear > used in his measurements (Spirent?, Ixia?, a school ruler?) should be the one > that claims the above mentioned fantastic figures.I'll take "fantastic" as a compliment. It's pretty trivial to reproduce and test my figures, but if you really don't believe me then how about this video, which shows the aforementioned 512MB test file being downloaded in a consistent 4.7s (measured using the 18Hz system timer tick and the "time" command, but you can easily check it yourself with a stopwatch): https://youtu.be/ERTdtxtCXWo If you want more proof, then the raw packet capture (first 128 bytes of each packet) is also available from: https://drive.google.com/file/d/0B1pvREz2MGuMc3ZfclJ5eFhMenc/view?usp=sharing Test setup here is: - Onboard Realtek 8168 NIC - Chainloading iPXE (bin/realtek.pxe) - 512MB test file pregenerated from /dev/urandom - Script does a repeated "imgfetch" to retrieve the 512MB test file - Directly connected via Cat 5E to my laptop, running Apache The whole setup is pretty vanilla. Nothing has been specially performance-tuned, and the test hardware is all at least five years old. Please ask if you want any more information. Michael