180 likes | 401 Views
Project – C IaaS – Resource Management. Akshay Patekar Nikhil Bharadwaj Ramesha Sharath Chandra Kanaka R Suthan Venkataramanaiah Vidya Ravivarma. AGENDA. Project Overview Background Study Architecture Q Clouds (Paper Exploration) Openstack Setup Ganglia Setup Instance Management
E N D
Project – CIaaS – Resource Management AkshayPatekar Nikhil Bharadwaj Ramesha Sharath Chandra Kanaka R Suthan Venkataramanaiah VidyaRavivarma
AGENDA • Project Overview • Background Study • Architecture • Q Clouds (Paper Exploration) • Openstack Setup • Ganglia Setup • Instance Management • Algorithm Design • Dynamic VM Resize and Live Migration • Issues faced • What Next?
Project Overview • Continuously monitor the cloud environment • Process the input and take decisions based on resource utilization metrics • An algorithmic approach • Perform load balancing • Dynamically adjust resources and live migrations of VMs
Background Study- Architecture Openstack Master Ganglia Master Gmond collects the monitoring data and sends it to the Gmetad which is running on Ganglia Master VM VM VM VM It directs VM s to migrate to another physical server in case of application degradation Load Balancing algorithm runs here on top of Ganglia monitored data and sends the decisions for VM migrations. A programs continuously runs by polling for the decisions from the Load Balancing Algorithm and calls upon Openstack APIs for VM migrations
Background Study - QClouds • A QoS aware framework that tunes resource allocations to mitigate performance interference effects • Three main Components in this framework • Scheduler : Module which polls and feeds ganglia monitored data to resource mitigation control • Resource Mitigation Control: Core module to arrive at optimized resource allocation values for performance degraded VMs • Dynamic Migrator: Module in charge of dynamic VM migration based on optimized allocation values • VMs are migrated based on Q-State • A custom performance metric • Our algorithm is based on this paper
Openstack Setup • Setting up the environment for Openstack • Openstack – multinode installation • Openstack master installed on host : Openstack • Openstack slave installed on host Compute1 and Compute2 • Installation done using devstack • $ >./stack.sh • Shell script is used for installing Openstack
Openstack – Modules • Modules configured • Nova, Keystone, Glance, Horizon • In Openstack, each VM has two IPs: fixed and floating • VMs are configured using these IPs • Nova, the compute component allows to manage multiple VM instances • Nova Manage is used to manage Openstack services • $ nova-manage service list • $ nova-manage service disable host=<host> service=<service_name>
Ganglia Setup • Ganglia master is run on a dedicated VM • Gmetad, Gmond , Ganglia web frontend installed and configured • Ganglia slaves (Gmond) run on each VMs • VMs running benchmark application • Configuration to specify Gmond representative • $> data_source “Cluster openstack" 192.168.0.131 • Master polls to fetch data from this Gmond • Metrics monitored are CPU_module, Disk_module, Load_module, Mem_module, Proc_module. • rrd file created for each of the metric
VM Instance Creation • Using nova command we instantiated VM on compute nodes • First we need to create key-pair for each compute node nova keypair-add --pub_key ~/.ssh/id_rsa.pub KEY_NAME • Creating Instance nova boot image <image name> flavor <flavor id> key-pair <keypair>
Algorithm Design GANGLIA MASTER OPENSTACK MASTER GMETAD VM1 VM2 VM3 Data Reader() Nova Module APIs DecisionMsg() Load Balancer() DecisionPolling()
Dynamic Resize • Nova resize command used to change the flavor of the running VM • nova resize <instance name> <flavor> • Flavor gives the new set of metrics for the VM • VM is resized while it is running • VM resize fails if it is in any other state • Nova Client Python used to write the python script to do resize • The script accepts instance to be resized and changes the flavor of the running VM • Our Load Balancing algorithm uses this script for dynamic resizing
Dynamic Resize Code This script takes one parameter – Source VM Name From novaclient.v1_1 import client from credentials import get_nova_creds import os import sys import time creds=get_nova_creds() nova=client.Client(**creds) instance_name=sys.argv[1] instances= nova.servers.find(name=instance_name) print instances.name flavor=instances.flavor print instances.id,instances.status,flavor['id'] instances.reset_state(state='active') flavor_id=int(flavor['id'])+1 instances.resize(flavor_id) print "Resizing" time.sleep(50) print "Confirm resize" instances.confirm_resize() time.sleep(20) instances.reboot(reboot_type='SOFT') print "Resize Done!!!"
Dynamic Live Migration • Set up NFS server in controller and NFS-common slaves • Shared storage based live migration • Ex.: /opt/stack/data/nova/instances • Mounted on all servers for visibility across all compute nodes • Initially tested with nova command • nova live_migration <instance_id> <host_ip> • Migration managed by the controller • Using nova client python wrapper functions, migration script was written • Instance ID and destination Host will provided as the arguments • Migration managed by the Controller
Dynamic Live Migration This script takes two parameters - source VM Name and Destination_server from novaclient.v1_1 import client from credentials import get_nova_creds import os import sys import time creds=get_nova_creds() nova=client.Client(**creds) source=sys.argv[1] #instance_id=sys.argv[2] destination=sys.argv[2] instance=nova.servers.find(name=source) print "Migration started" instance.live_migrate(destination,False,False) print "Migration Done!!!"
Issues faced and solved • Not all Services come up after the machine is restarted • Need to run rejoin_stack.sh script • Nova network services enabling issue solved using NTP server • To keep controller and all compute nodes in synch • No proper documentation for usage of Nova Client python wrappers • Referred the source code to write the dynamic resizing and migration scripts
What Next? • Install benchmark applications and configure Ganglia for these instances • Gmond monitors the application resource usage • Design of load balancing algorithm based on Q-Clouds • Implement and Test the algorithm by using different sets of VMs in the Openstack cloud environment • Improve the algorithm to meet scalability