ToDo

A list of the future tasks to be implemented on HPC2, may be brief or somewhat in detail. Not-so-urgent types of tasks are listed in italic way.

Hardware Level

Network related

• Enable jumboframe in LAN

  • config on MTU of servers

  • config on MTU of switches, Jumbo

• Divide S1720 switch into two VLAN

• DHCP on the master node with very long lease time

Server related

• iDRAC configuration

• Open manage configuration

• include more old machines into the cluster

Software Level

OS related

• locale to en.US for all systems and timezone

• ntp server for master node

• nfs settings

• dhcp server for master node

• apt and pip source change

• ? shutdown password login in LAN

• user resource limit

  • disk limit in home

  • resource limit in master

  • qos in slurm

  • pam in slurm

• sshfs

• rclone

• rsyslog forward (by ELK stack)

• gitlab

• syncthing

DevOp softwares

• cobbler

• ansible

  • create playbooks for:

  • user management and their ssh key

  • apt tools install

  • service state management

  • network configuration (dhcp server with fix ip on master and dhcp auto on nodes and nat)

• ?nagios

• ganglia

• modules

• backup tools

HPC softwares

• spack

• slurm

• intel parallel studio (ifort icc intelmpi intelpython and mkl included)

• Eigen and armadillo

• boost

• GSL

• distributed mathematica

  • automatically activation

  • launch remote kernel via cli

• SLEPc PETSc and external linear and eigen solver packages

• Julia

• ipyparallel

• Matlab

Modern cloud and distributed system tools

• OpenStack

• Spark

• Kubernetes

• Elasticsearch

• ?ceph

Miscs

• opensource the ansible playbooks in this HPC

• ansible authorized key double check

• possible setup one more v2ray inbound which has only outbounds to 176 network, used for other users to access jupyter notebook

• spack install hpl for linpack benchmark(just use mkl one)

• backup mysql database

• ? backup elastic database

• bootstrap setup for new machines by curl scripts

• more careful division on playbooks, new roles comes in! gpu partition and shared storage on compute nodes, backup manage node (backup of slurmctld slurmdbd and possible elastic node)

• change mount logic to more robust and support on [sn]

• add a debug partition queue and a gpu queue for slurm

• replication of ES

• authetication of ES

• make hostname consistent by ansible on ubuntu18.04, (a detailed study on cloud init subsystem)

• ganglia incomplete metric collection

• ganglia gpu plugin

• elastalert for mail alerting

• MPI benchmark

• GlusterFS

tasks to explore

• unifomity of user

• uniformity of environment variables (propagate by sbatch)

warning

• name system with version number

• ?try install all useful things in opt or home which is going to be export

Future directions

• combination of k8s and slurm

• hybrid HPC/cloud setup and elastic scaling

design principle

• more on master, less on slave, everything in slave should be included in ansible workflows

• more on ansible playbooks, less by hand