[ipxe-devel] Win7 x64 iSCSI install - insanely slow shutdown time

Andrew Bobulsky rulerof at gmail.com
Tue Jun 4 16:18:53 UTC 2013

On Mon, Jun 3, 2013 at 7:38 PM, Steve Cross <hairlesshobo at stevecross.org> wrote:
> Hey everybody,
> Been a little while since I posted on here. I finally got my Intel
> NUCs that I wanted to get and they run great. Got them up and running
> with iSCSI and everything is great - except shutdown time. The system
> will go from power off to running windows in 60 seconds, but it takes
> 6-8 minutes to shut down and I have NO clue why. This is quite
> annoying because whenever I do have to shutdown my iSCSI server, I
> want to make sure that all clients are cleanly shutdown first and this
> is quite an insane length of time for a very clean install of Windows
> 7 to shut down. I was wondering if it could have anything to do with
> iSCSI, or if perhaps there may be some suggestions (not directly
> related to iPXE) that you might could offer.
> Thanks in advance!
> Steve Cross
> hairlesshobo at stevecross.org

Hello Steve,

During Windows shutdown, aside from general write cache buffer
flushing, sometimes the process of reconciling the pagefile can slow
things down, too.

I can suggest that, if you really want to debug this issue from the
Windows side, you could run Sysinternals' Process Monitor... though it
occurs to me that while the tool supports boot-time monitoring in
addition to running on a live system, I have no idea how it would play
out for shutdown.

As a general rule, I like to disable the swap file on iSCSI-booted
systems, or at least move the page file and hibernation files to a
local disk.

You can disable the swap file from the GUI as normal, and the
hibernation file can be shut off by running

C:\> powercfg /h off

from an elevated command prompt.

This may not fix your problem at all, but I thought I'd throw in my
two cents as to where I would start, at least on the Windows side.

Additionally, on the iSCSI target, make sure it's configured with at
least a reasonable amount of write-back cache; say a gig or so.  That
really speeds up buffer flushing on the initiator side (or I'm
suffering from the placebo effect, but I swear it works :P).

Andrew Bobulsky

More information about the ipxe-devel mailing list