Environment

Endpoint Central 11 (build 11.5.2624.08) on a Windows Server 2022 VM.

Two NICs: one for agent communication, one dedicated 10 Gbit vmxnet3 NIC in an

isolated deployment VLAN. DHCP is served by a FortiGate (next-server + filename

for UEFI x64). PXE media is WinPE 11 x64 with a Dell driver CAB, boot.wim ~500 MB.

Targets are physical Dell machines, UEFI only.

Problem

bootx64.efi loads fine. boot.wim averages 8-14 Mbit/s, so 7-10 minutes per machine.

The throughput graph shows it is not uniformly slow: it bursts up to 54 Mbit/s,

then drops to zero, about ten times per minute. The average comes from the dead

time between bursts, which looks like packet loss and retry waits rather than a

throughput limit.

We tested BLKSIZE 1456 / 16384 / 32768 and WINDOWSIZE 4 / 8 / 16, always changing

tftpSettings.conf and the matching BCD ramdiskoptions together and restarting the

PXE service. Peaks change, the average does not. We are already on the 16384/16

values recommended in the thread "PXE boot.wim files takes forever to load".

TIMEOUT=3000 aborts the transfer entirely. Only one process holds UDP 69, firewall

rules are in place, server CPU stays below 15%.

Questions

1. What causes the repeated drops to zero? Known issue when the server NIC is

   10 Gbit and clients sit on a 1 Gbit segment?

2. What does WINDOWSIZE_OPTIONS_ENABLED do and which values are valid? Undocumented,

   and OSD_size_updation.bat always forces it to 2.

3. What unit and range does TIMEOUT use?

4. TFTP_HOSTED is set to the deployment NIC, but netstat shows 0.0.0.0:69. Can the

   PXE service be bound to one interface?

5. Any documentation for a central server with two NICs, separating agent traffic

   from PXE/OSD traffic?