fabricattachedmemory / emulation Goto Github PK

the latest release of vmdebootstrap (1.6 as from here: https://vmdebootstrap.alioth.debian.org/doc/overview.html) contains the following code:

        if self.settings['image'] and self.settings['tarball']:
            raise cliapp.AppException(
                'Use --image or --tarball, not both.')

which causes the emulation_configure.bash script to fail in line 300ff:

    quiet $VMD $VAROPT --log=$LOG --image=$TEMPLATE --tarball=$TARBALL \
        --mirror=$MIRROR --owner=$SUDO_USER

When I run emulation_configure.bash it gives an error when trying to create the node_emul network, claiming that the network is already in use by interface fabric_emul. I downloaded Emulation_master.zip on three days ago, on June 26, 2017. It looks like fabric_emul was in an earlier version from March. How do I get rid of fabric_emul without reloading Linux and starting over?

[2017.06.29 09:47:17 virsh 4062] INFO commands: "connect"
[2017.06.29 09:47:17 virsh 4067] INFO net-list: all(bool): (none)
[2017.06.29 09:47:17 virsh 4071] INFO net-destroy: network(optdata): node_emul
[2017.06.29 09:47:17 virsh 4071] INFO net-destroy: found option : node_emul
[2017.06.29 09:47:17 virsh 4071] DEBUG net-destroy: trying as network NAME
[2017.06.29 09:47:18 virsh 4071] ERROR Failed to destroy network node_emul
[2017.06.29 09:47:18 virsh 4071] ERROR Requested operation is not valid: network is not active
[2017.06.29 09:47:19 virsh 4076] INFO net-undefine: network(optdata): node_emul
[2017.06.29 09:47:19 virsh 4076] INFO net-undefine: found option : node_emul
[2017.06.29 09:47:19 virsh 4076] DEBUG net-undefine: trying as network NAME
[2017.06.29 09:47:20 virsh 4081] INFO net-define: file(optdata): node_emulation.net.xml
[2017.06.29 09:47:20 virsh 4085] INFO net-start: network(optdata): node_emul
[2017.06.29 09:47:20 virsh 4085] INFO net-start: found option : node_emul
[2017.06.29 09:47:20 virsh 4085] DEBUG net-start: trying as network NAME
[2017.06.29 09:47:20 virsh 4085] ERROR Failed to start network node_emul
[2017.06.29 09:47:20 virsh 4085] ERROR internal error: Network is already in use by interface fabric_emul

Dear all,

Actually, I have some questions which are related to FAM in the virtualization context and they are as follows:

• I want to know is it really necessary to virtualize FAM?!(We know that this project is an emulation to mimic the FABRIC ATTACH MEMORY behavior because the hardware is not available to everyone right now then we should emulate, maybe some people have access to this hardware but I do not :) )

• If we could build a system that could virtualize FAM then who cares about that?! (I think to myself how virtualization originated from the 60s for the reasons such as multiprogramming and high utilization and the high price of hardware and it has been keeping up with the cloud technologies and HPC and so). But today do we have the same aims to build a system for that?!
  I guess if we could find sensible reasons to virtualized FAM then next step would be looking for current hypervisor support requirements and try to develop them with respect to the requirement and limitation. for instance, Is it possible for the current memory management components of hypervisors to support this amount of memory available in FAM

• What are the probable use cases of virtualized FAM?

• And there are more challenges that could be discussed in this context but the question is to find reasons for FAM virtualization at first place.

Thank you

AfterI install FAME and VMs seem to run properly,
we still annoyed to share memory from several processes on several nodes.
Though we saw software stack slide from THe Machine presentation,
but it's hard to be corresponded on to one.
Could you give us what is(are) the minimum component(s) to test, for example which to be installed to run Sparkle?
We would like to know which software and libraries are required to share memory from multiple nodes.
And also which APIs or functions should be call from processes.
We couldn't find any documents in github and these information will help users who try to test FAM(E).

Is there any Benchmark or test suite through which we can evaluate FAME in terms of performance, latency and throughput?
And which module in this emulation are implementing Gen-Z memory controller?

Thanks in advance for your time and response.

I am trying to run the FAM emulator on Ubuntu 16.04. Here's what I did to get a clean run of emulation_configure.bash.

Missing dependencies installed with "sudo apt-get install":
zlib1g-dev
gtk2-dev
libglib2.0-dev
dh-autoreconf
bison
flex
vmdebootstrap
libvirt-bin

I installed qemu 2.8.1.1 from source.

I made two minor changes to emulation_configure.bash:

1. add /usr/local/bin to PATH (for qemu-2.8.1.1)
   export PATH="/bin:/usr/local/bin:/usr/bin:/sbin:/usr/sbin"
2. make 2.8.1.1 a verified qemu version
   declare -a VERIFIED_QEMU_VERSIONS=("2.6.0" "2.8.0" "2.8.1.1")

Handle VM golden image /tmp/node_template.img

Creating new /tmp/node_template.img from http://ftp.us.debian.org/debian
CRITICAL:root:Traceback (most recent call last):
File "/usr/lib/python2.7/dist-packages/cliapp/app.py", line 179, in _run
self.settings.load_configs()
File "/usr/lib/python2.7/dist-packages/cliapp/settings.py", line 729, in load_configs
s = self.set_from_raw_string(name, value)
File "/usr/lib/python2.7/dist-packages/cliapp/settings.py", line 704, in set_from_raw_string
s = self._settingses[name]
KeyError: 'sparse'

Traceback (most recent call last):
File "/usr/lib/python2.7/dist-packages/cliapp/app.py", line 179, in _run
self.settings.load_configs()
File "/usr/lib/python2.7/dist-packages/cliapp/settings.py", line 729, in load_configs
s = self.set_from_raw_string(name, value)
File "/usr/lib/python2.7/dist-packages/cliapp/settings.py", line 704, in set_from_raw_string
s = self._settingses[name]
KeyError: 'sparse'
Usage: vmdebootstrap [options]

vmdebootstrap: error: no such option: --debootstrapopts
sudo -E chown jwilkes /tmp/node.*
chown: cannot access '/tmp/node.*': No such file or directory
Error: Build of /tmp/node_template.img failed