Re: om-2.6 patch for kcomd (pre-pre-alpha)

DOH! Accidentally hit send.  Sorry 'bout that.

I've put my stuff on 
http://www.osource.org/openmosix/om-26.html

The patch is against the vanilla 2.6.16 instead of against git.

I've tested the patch on there now, om-2.6-full-20060714.patch.bz2, and
it applies, compiles and runs.

Matt

On Fri, 2006-07-14 at 16:56 +0000, Matt Dew wrote:
> Hi again,
>  As has been pointed out to me that patch applies but does not
compile.
> I have done git pulls periodically to see if there are any changes but
I
> have gotten '...up-to-date...' messages so I assumed that meant there
> were no changes.  Unfortunately there were, so that patch will not
work.
> 
> 
> 
> 
> 
> 
> Tab,
> 
> When using the straight git, if kcomd is compiled as a module, the
> following error pops up.
> 
> hpc/built-in.o(.text+0xe57): In function `comm_accept':
> hpc/comm.c:296: undefined reference to `sock_alloc'
> make: *** [.tmp_vmlinux1] Error 1
> 
> It's due to sock_alloc being declared static (KCOMD_NSTATIC) unless
> kcomd is selected as 'y'.  
> 
> Attached is a patch to let it compile if kcomd is a 'm' or a 'y'.
> 
> 
> 
> 
> 
> On Wed, 2006-07-12 at 01:25 +0000, Matt Dew wrote:
> > Hi all,
> >   Here's a patch to the current git tree (7/11/06) that contains the
new
> > kcomd code.   kcomd is the new kernel daemon that handles migration
for
> > the processes.
> > 
> > Moshe, Bruce, if this patch is too large I can put it on a web
server
> > for people to download.   ?
> > 
> > There is a README included but here's a quick intro.
> > 
> > All communication between nodes is done via packets.  Each packet
> > consists of a header, which has information about which pid this
packet
> > belongs to, how much data accompanies this packet, the type of
packet
> > (initial migration, system call, signal, etc), and the payload of
data
> > that immediately follows.  
> > 
> > The processes are responsible for generating the packets, adding
them to
> > linked lists and then signalling the kcomd daemon that a packet is
ready
> > for transmission.  All this is done in the kernelspace stubs of the
> > userspace processes.  The userspace programs don't see anything.
> > 
> > kcomd sleeps until there is work to do. Either a received packet to
> > process or a packet to send.  Then goes back to slees.
> > 
> > kcomd wakes up on the signal and sends the packet to the
corresponding
> > node.  When a packet is received, kcomd wakes up and checks to see
which
> > process this packet belongs to, puts it into the process's linked
list
> > and informs the process it has either new data or a response to a
> > previously sent packet.
> > 
> > The code isn't very clean but if I wait until I'm happy with it I'll
> > never send it. :)
> > 
> > compile kcomd as a module.
> > 
> > Currently, issuing a migration command (echo $IP >
/proc/../om/where)
> > will send a migration init packet to the other node, which sends a
> > reply.   You can run as many simulaneously as you want.  There
doesn't
> > appear to be any memory leaks. 
> > 
> > This is just to show that communications are working and the
foundation
> > plan is solid.  Adding the code for real migrations is next, as is a
> > massive cleanup.  :)
> > 
> > Please try the patch and review the code. I'm not a programmer and I
am
> > asking for help.  There are plenty of commented printk statements
you
> > can uncomment to get a better idea of what's happening.
> > 
> > 
> > config options:
> > #
> > # HPC options
> > #
> > CONFIG_KCOMD=m
> > CONFIG_OPENMOSIX=y
> > # CONFIG_OPENMOSIX_VERBOSE is not set
> > # CONFIG_OPENMOSIX_DEBUG is not set
> > # CONFIG_OPENMOSIX_CTRL_FS is not set
> > 
> > Matt



-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642