Transcript david

Construction methods and
monitoring in meta-cluster
systems
Korenkov V.V, Mitsyn V.V, Chkhaberidze D.V, Belyakov D.V
LIT, JINR
Main goals




To unite in city the distributed resources in
one computing system (meta-cluster)
Installation of detailed monitoring of the
system,
Develop and debug convenient, fast
mechanisms for expansion of the cluster
To make management and administration of
the system centralized and to develop ways
of automation for some procedures
Main problems




presence of heterogeneous platforms in the
rectangular components
Addition of cluster-node facilities on the PCs
should not prevent their main usage
Resources are distributed in different places
over the city
some functions and software on the host
operating system impends the work of
system under virtual pc
Loading process
PC emulated
by VMware.
SERVER
dhcp
tftp
nfs
ntp
openbps
warewulf
Gagnlia srv
Get IP address
Load Kernel and file
system
Mount some repositories
Synchronize time
Get Jobs
Real
windows
Logical structure
Virtual
Primary Out,
Internet
Bridge
University
SERVER
Real,
Windows
eth0
Schools
eth1
Cisco
eth2
Out, With Nat
Cluster map
Server
Used Software






CERN Scientific Linux 3.0.5
VMware – virtual PC simulator
VLan – Virtual Network Simulator
Warewulf – technology for creating and
booting cluster nodes from network without
Hard Disks.
OpenPBS – Portable Batch System
Ganglia Monitoring System
Problem Solving



By means of VMWARE, for creations of
system we receive a homogeneous
environment from heterogeneous platforms
VMWARE in WINDOWS starts as
background process with a priority one less
than usual (Bellownormal)
Expansion of resources occurs using of
cloning of a hard disk from an exemplary
computer
Monitoring



Was used Ganglia cluster monitoring system
Work nodes are incorporated in different
groups on classes. One class – one group
On a site it is added an opportunity for
detection of mistakes from the point of view
of windows
Screenshot from server
Screenshot from Check errors
How “check errors” works
SERVER
PING
Virtual,
VMware
eth1
PING
eth2
Real,
Windows
Results



For creation metacluster it was used only
existing technologies
Computing resources have not been bought,
the system is created from already being
computers
Use of computers as computing units of the
system did not hinder their main functions
Results


Heterogeneity of platforms is endowed by
means of VMWARE
Monitoring does administration of the cluster
more convenient and helps to support
reliability of the system.