Differences

This shows you the differences between two versions of the page.

--- cluster:132 [2014/07/10 14:11]
hmeij [LXC Linux Containers]
+++ cluster:132 [2014/08/11 14:08] (current)
hmeij [Tests]
@@ Line 30: / Line 30: @@
-Ah, I'm simply going to CentOS7 because LXC is fully integrated.
+Ah, I'm simply going to CentOS-7 because LXC is fully integrated
+  * [[http://www.theregister.co.uk/2014/07/07/centos_7_gm/]]
+Ah, no.  Took a look at v7 and it's very different with the over engineering part of system control.  Will learn later, back to v6.5. I also want to preserve the ability to run different OS flavors which is not possible under LXC containers.
+==== KVM ====
+KVM is hardware level virtualization as opposed to my previous meanderings into the world of software level virtualization (Xen [[cluster:127|Virtual HPCC services]]).  There is basically a "dom0" domain manager running keeping track of virtual images that run from block devices written onto a block device.
+  * Followed the primary setup on this page
+    * http://www.howtoforge.com/virtualization-with-kvm-on-a-centos-6.4-server
+  * Once ''virt-manager'' worked we followed this doc a bit setting up bridges etc
+    * http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=2&ved=0CDQQFjAB&url=http%3A%2F%2Flinux.dell.com%2Ffiles%2Fwhitepapers%2FKVM_Virtualization_in_RHEL_6_made_easy.pdf&ei=GZfjU9P2KqS-sQS5p4HgCw&usg=AFQjCNFgCfRXV-g4eGE_Fm5AiIw4thnfYg&bvm=bv.72676100,d.cWc&cad=rja
+      * for future reference, more details
+        * http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&ved=0CC4QFjAA&url=http%3A%2F%2Flinux.dell.com%2Ffiles%2Fwhitepapers%2FKVM_Virtualization_in_RHEL_6_Made_Easy_Part2.pdf&ei=GZfjU9P2KqS-sQS5p4HgCw&usg=AFQjCNGzBgZ3f-YvCbV5sav7450BX_Io6Q&bvm=bv.72676100,d.cWc&cad=rja
+  * But once I got to the cloning part, I found this to be super easy and scritpable
+    * http://rwmj.wordpress.com/tag/virt-clone/
+First I create my v1 clone with virt-install then build my v1 clone the way I wanted it and then customized it manually (/etc/fstab, openlava scheduler, /etc/passwd, /etc/shadow, /etc/group and /etc/hosts, etc). Added a second bridge (''br1'' is public/internet, previously created ''br0'' is for private network for scheduler).  That was done via virt-manager GUI (View VM Details of vm and then select Add Hardware). ''br1'' allows me to run yum update before I clone. Power this clone up, add to a test queue, and submit some jobs to make sure it all works.  Then clone.
+<code>
+# create v1, my "base" clone: 8 gb hdd, 1 gb ram, centos 6.5, connect to ethernet bridge ''br0''
+virt-install --connect=qemu:///system -n v1 -r 1024 --vcpus=1 \
+--disk path=/var/lib/libvirt/images/v1.img,size=8 \
+-c /var/lib/libvirt/images/CentOS-6.5-x86_64-bin-DVD1.iso \
+--vnc --noautoconsole --os-type linux --os-variant rhel6 --accelerate \
+--network=bridge:br0 --hvm
+</code>
+Here are the steps for cloning v1 to v5. You'll need ''guestfish'' part of the libguestfs tools allowing you to edit content inside the block level file.  Very nifty.
+  * yum install libguestfs-tools-c
+Next we'll create the v5.img block device, dump the v1 config into v5.xml and then edit that file. UUID and Mac Address we'll edit and the last 2 characters we change to '05' (make it unique). Name becomes v5 from v1 and we'll similarly adjust the block device location/filename. Then we define this as domain v5.
+Launch ''guestfish'' and make the same edits to Mac Address in files listed below. Then boot the vm.
+<code>
+# cd /var/lib/libvirt/images
+# dd if=v1.img of=v5.img bs=1M
+# virsh dumpxml v1 > /tmp/v5.xml
+# vi /tmp/v5.xml
+(change name, uuid, file location, mac address(es))
+# virsh define  /tmp/v5.xml
+Domain v5 defined from /tmp/v5.xml
+# guestfish -i -d v5
+Welcome to guestfish, the libguestfs filesystem interactive shell for
+editing virtual machine filesystems.
+Operating system: CentOS release 6.5 (Final)
+/dev/vg_v1/lv_root mounted on /
+/dev/sda1 mounted on /boot
+><fs> edit /etc/sysconfig/network
+(change hostname)
+><fs> edit /etc/sysconfig/network-scripts/ifcfg-eth0
+><fs> edit /etc/sysconfig/network-scripts/ifcfg-eth1
+(for nics comment out any UUID lines, change hardware address, change IP)
+><fs> edit /etc/udev/rules.d/70-persistent-net.rules
+(change hardware address)
+><fs> exit
+# virsh start v5
+Domain v5 started
+</code>
+==== Tests ====
+^ Melt Lammps LJ problem, 10000 steps with 32000 atoms Loop Times (secs) ^^^^^^^^^
+|  2  |  4  |  6  |  8  |  10  |  12  |  14  |  16  |  nr of jobs  |
+|  556  |  560  |  588  |  625  |  760  |  936  |  1122   |  1300  |    |
+|  (linear)  ||||  781  |  938  |  1094  |  1250  |    |
+|  (n35:load 32)  ||||  726  |    |    |    |    |
+|  (hyperthreading)  ||||    |    |    |  ????  |  16 logical cores  |
+VMs running in queue ''testvm'' on dual quad core PE2950 (8 cores, 16 GB ram)
+  * First ran on n35 (32 core node under hyperthreading will full load) with an average loop time 726 secs.
+  * As I submit jobs to the VMs they perform well up to 8 jobs (one job per core; dual quad core node).
+    * That is with the KVM overhead
+  * Assuming a linear penalty for over committing, 16 jobs is expected to take Loop times of 1250 secs.
+    * However after 10 jobs we're surpassing that penalty threshold
+  * And then I was to turn on Hyperthreading creating 16 logical cores
+      * To my dismay this chipset does not support that, bummer!
+      * Was expecting to gain some performance back ...
+      * Maybe try on newer hardware when idle ...
+But we learned Xen and KVM setups, and
+  - We can now support a heterogeneous environment if we wanted (Suse, Scientific Linux, Windows (eh, what?))
+  - Use a KVM environment up to the number of cores on a box without penalty
+  - And change the mix of nodes if needed (more/less cores per node, memory size etc)
+  - Still, not an answer for my "high core count/low memory footprint" problem, see [[cluster:133|high core count/low memory footprint]]
 \\

DokuWiki

User Tools

Site Tools

Differences

Page Tools