Resurse fizice - RaaS-IS

Resursele fizice ale centrului de cloud privat RaaS sunt formate dintr-o colecție de două sisteme blade HPE Synergy 12000 interconectate Exthernet prin topologie spine-leaf prin multiple link-uri de 100Gbps.

Puterea de calcul este asigurată de 16 servere HPE Synergy 480 Gen10, cumulând 576 core, respectiv 1152 HT, la o frecvență de 2,6 GHz, aproximativ 40 TFLOPS per CPU. Memoria RAM cumulată este de 2 TB la frecvență de 3 GHz. Fiecare server este dotat cu stocare locală redundantă: 300 GB SAS la 12 Gbps. Stocarea locală este destinată doar sistemului de operare. 

Conectivitatea serverelor este de 25 Gbps (dual) în LAN și 32 Gbps multipath către SAN.

Sistemul de stocare este un multi-controller cu interfețe 32 Gbps FC multipath si cache de 128GB, asigurând: stocare rapida 10 TB prin SSD sas la 12Gbps per disk și stocare lenta aproximativ 0.5 PB (petabyte) prin NL-SAS la 12Gbps per disk. Fluxul de scriere asigurat de combinația de SSD și SAS este de peste 150.000 IOPS (operații scriere/citire pe secundă).

Întreaga infrastructură fizică este protejată la căderile de curent printr-un UPS de EATON, care asigură la încărcare optimă, timpul necesar pentru oprirea în siguranță a echipamentelor. Pentru întreruperile care nu depășesc anumită continuitatea funcționării echipamentelor este asigurată de generatorul din dotarea Direcției de Comunicații Digitale din cadrul UAIC.

​ ​Power computation of RaaS-IS	​ ​Storage of RaaS-IS
​ For computation: 2 x blade system HPE Synergy 12000 Frame Full redundancy: management blade system, fan, power, networking and fibre channel 16 x compute node HPE Synergy 480 Gen10:  2 x Intel Xeon-Gold 6240 (2.6GHz/18-core)  128 GB RAM @ 2933 MHz 2 x 300GB SAS 12G Enterprise 10K in RAID 1 2 x 25 Gbps Ethernet and 2 x 32 Gbps FC For management: 4 x rack server HPE ProLiant DL360  2 x Intel Xeon-Gold 6240 (2.6GHz/18-core)  128 GB RAM @ 2933 MHz 2 x 300GB SAS 12G Enterprise 10K in RAID 1 2 x 25 Gbps Ethernet and 4 x 1 Gbps Ethernet	​ Enterprise storage system HPE 3PAR 8440 2N hybrid system with SSD and HDD flash optimization architecture with 64 GB Cache 18 x 920GB SAS SFF (2.5in) SSD in RAID 5 96 x 8TB SAS 7.2K LFF (3.5in) HDD in RAID 6  tier configuration for optimization SSD-> HDD (with statistics of data write/read) Performance  162K IOPS Mixed (read/write), 110K write 3.5 GB/s read, 2.4 GB/s write Capacity 760 TB RAW, 550 TB USABLE IO 4 x 32 Gbps FC and 4 x 16 Gbps
​ ​Network and fiber channel	​Hardware management and OS ​
​ Ethernet Switch in blade sistem 4 x Mellanox SH2200 TAA with  12 x 25 Gbps downlinks to compute module ethernet adapters 8 x Quad-Small Form-factor Pluggable 28 (QSFP28) external uplink ports configurable as 1 x 100 Gbps  Top of the rack switch 2 x Aruba 8325-48Y8C with 48p 25Gbps  and 8p 100Gbps Topology configuration: full redundancy in spine leaf architecture Fibre channel Switch in blade sistem 4 x Brocade 32Gb/20 4SFP+ with  12 x 32 Gbps downlinks to compute module FC adapters 8 x 32 Gbps external uplink ports Top of the rack switch 2 x Aruba 8325-48Y8C with 32 Fibre channel ports capable of speeds of 8/16/32 Gbps  Topology configuration: multipath fibre channel Redundancy	​ Metal-As-A-Service (MAAS) provisioning with Linux (Ubuntu) Infrastructure monitoring and discovery Ansible, Chef, Puppet, SALT, Juju integration Disk and network configuration API-driven DHCP, DNS, PXE, IPAM Hardware testing and commissioning
​ ​Cloud implementation	Openstack services and placement ​
​ Openstack was deployed using Juju as the automation tool Juju is an open source application modelling tool that allows you to deploy, configure, scale and operate cloud infrastructures quickly and efficiently on public clouds A Juju Charm is a structured bundle of YAML configuration files and scripts for a software component that enables Juju to deploy and manage the software component as a service according to best practices. Charms are composed of metadata, configuration data, and hooks with some extra support files. Openstack is deployed in a High Availability configuration, which means that for each service that needs to be HA, 3 different units of the same type of service are deployed on the 3 controllers.  There are also services that have HA built in. These services are: ceph-mon, rabbitmq-server and mysql-innodb-cluster. These services do not use pacemaker/corosync for the HA, but use haproxy for loadbalancing.	​ Main Barbican Cinder Glance Heat Keystone Magnum Neutron Nova Horizon Placement Sahara	​ Shared  Ceph EasyRSA ETCd Memcache Mysql NTP OVN RabbitMQ Vault	​ Monitoring Grafana Prometheus Telegraf FortiSIEM