[gPXE] Transfer speed with e1000 and pcnet32 in VMware

[Stefan Hajnoczi]

Thanks for uploading the packet capture file.  There are ~5 sec stalls
at 5.8s, 23.4s, and 41.6s.

The pattern I see before each stall is:
3 ACKs from gPXE -> web server.  All received data acked.
Stall for several seconds.
Web server transmits data again.

Can't see any issues with TCP windows, looks like both ends have window space.

I believe gPXE has done its part and we are waiting on the web server
when the stall occurs.

Can you describe the networking setup between the gPXE VM and the HFS
web server?  What OS and version is the web server running?

I tried fetching a 100 MB file via HTTP over the internet from a Linux
Apache web server using QEMU and the e1000 driver.  I was unable to
reproduce the stalls - that's not a surprise since my setup differs
from yours in several ways.

Are you able to test Virtualbox or QEMU?  Both can do e1000 hardware
emulation.  Performing this test might help us determine whether this
is VMware-specific.

Has anyone else analyzed the packet capture Joakim uploaded?

Stefan