Cloud Pak for Data System overview

Cloud Pak for Data System is an all-in-one cloud-native Data and AI platform in a box, that enables you to collect, organize and analyze data with unprecedented simplicity and agility, within a preconfigured and governed environment. Combining storage, compute, networking and software into plug-and-play nodes, this Intel x86-based hyper-converged infrastructure helps you accelerate private cloud deployment to a matter of hours. It can easily and elastically scale to suit your changing data and AI needs. With Cloud Pak for Data System, software and system management is simplified through a single intuitive dashboard and you benefit from a flexible pay-as-you-go capacity model.

Cloud Pak for Data System provides a modern private cloud experience. It helps you align your data center strategy with your could strategy - you can modernize your environment without moving to public cloud with the following features:
  • All-in-one, pre-integrated system that consists of sophisticated hardware and software ready to go from day 1 operations, and is optimized for data and AI workloads.
  • Portability and scalability with Red Hat OpenShift - you can run the workloads consistently like in a public cloud.
  • You can securely store and process data locally to leverage the AI ladder with Cloud Pak for Data: collect, organize, analyze data and infuse AI
  • The system infrastructure provides an active-active 25Gb HA network and is 50x faster than the public cloud.

Hardware

Cloud Pak for Data System hardware comes in different sizes. You can have the hardware components installed in your own rack, or go with the integrated rack solution. You can also expand the base system with additional enclosures, redundant management and fabric switch, or even a spine switch. For more information, see Expanding Cloud Pak for Data System.

Cloud Pak for Data System chassis is based on Lenovo ThinkSystem SD530. The base configuration consists of two 2U Lenovo D2 enclosures, each containing four front-access SD530 servers (nodes). The enclosures are connected with a single 25G fabric switch and a single 1G management switch.

Each SD530 server is a node. Each node contains four NVMe drives and the capacity of these drives defines the type of the node.

Table 1. Node type and its capacity
Enclosure type Node Type Raw Data Storage
Large Large 64 TB
Node specifications:
Server type SD530
CPU 16 cores 2.1 GHz; SMT=2 Enabled
Note: Hyperthreading is enabled by default on x86 processors: each has 1 physical core and 1 hyperthreaded.
Total memory 192 GB
Memory DIMM 16 GB 2666 MHz DDR4
Management Network Connection 2 x 10G RJ-45
Fabric Network Connection 2x 25Gb/s Mellanox Innova-2 SmartNIC
Internal drives

Large: 4 x 4 TB NVMe  
Note: Depending on the vendor, the NVMe drives might be 3.84 TB or 4 TB.
Internal M.2 HBA Marvell 88SE9230
Internal M.2 drives 2 x 480 GB SATA

System nodes

Physical nodes are contained in enclosures. Each enclosure contains four nodes. Nodes are assigned roles. The roles are implemented by deploying one or more virtual machines on a server node.

The bare metal servers in the system host bare metal services and one or more virtual machines. Nodes can be assigned one of the following roles:
Control
Three servers in a system are always assigned a control role and these three nodes are designated as a Control Plane. These servers are candidates for a Platform Manager hub. They host one Control Virtual Machine.

Control nodes manage your cluster and your Cloud Pak for Data System deployment. At least two of the three control nodes must be operational for the system to work.

Worker
Compute nodes. They can host one or more Cloud Pak for Data System Worker Virtual Machines. Worker nodes run the services in your cluster. There are two types of worker nodes:
  • Universal: Can host any service, container, or pod as designated by Cloud Pak for Data System.
  • Labeled: A dedicated VM to host only a specific pod or application designated by "Label".

When a new process starts, the control plane determines which node has sufficient capacity to run the process. Cloud Pak for Data System can continue to run when multiple worker nodes fail. However, you might notice that performance decreases when multiple nodes are down.

Additionally, if a node fails, the control plane attempts to bring any active processes up on another node. While the control plane attempts to bring up the processes, you might experience an outage.

Software

Red Hat Operating System
  • World leading enterprise Linux platform
  • "Military-grade" security
  • High performance, high uptime
Platform Management
  • System management
  • System monitoring
  • Events and alerts
Platform Support Elements
  • HPI - Host Platform Interface
  • CallHome
  • Resource Managers
  • ApUtil - Command line utilities to monitor and support the system
  • ApComm - platform communications
  • ApStor - Configuration and monitoring of platform storage resources
RedHat OpenShift
  • Comprehensive enterprise-grade application platform
  • Built for containers with Kubernetes
  • "Build, deploy, and scale."
Cloud Pak for Data System Edition
You can verify which version is included in What's new in 2.0.
  • Data and analytics platform
  • Built-in governance
  • Collect, organize, analyze data and infuse AI
Netezza® Performance Server for Cloud Pak for Data System
Available on extended systems, that is, systems with additional enclosures, as an optional service (add-on)
  • Data warehouse solution based on the same code as Netezza
  • Fully compatible with Netezza appliances
  • Optimized for high performance analytics with built-in hardware acceleration

Network

Base system network components:
Fabric Switch (8831-25C)

Mellanox SN2410 is a 48-port 10/25 switch with 12 additional ports that can run at 10/25/40/100. It is equipped with redundant hot-swappable fans and power supplies. The switch is running Cumulus Linux.

Data Fabric Network is the active-active 25Gb connections to the Mellanox Innova-2 SmartNICs. Innova-2 is used for Netezza and Natural Language Processing acceleration.

Management Switch (8831-S52)

Edge-core AS4610-54T-O-AC-B GigE switch (8831-S52) consists of 48 x 1G RJ-45 and 4 x 10G SFP+ and 2 x 20G QFSP+ stacking ports. It is equipped with redundant hot-swappable power supplies and redundant (not hot-swappable) fans. The switch is running Cumulus Linux.

The RJ-45 connections are to the EIOM card on the enclosure. The EIOM 1Gb connections are reserved for the management network.

For more information on network configuration, see Network configuration.

Storage

For information on storage, see Platform storage.

Security

Cloud Pak for Data System is and air-gapped solution.
  • Encryption scheme where data requiring protection is transformed into an unreadable form
  • Data encryption with built-in cryptographic algorithm and encryption key
  • Self-encrypting SSD disk storage
  • Compliance with high-security protocol - FIPS/STIG
  • Cloud Pak for Data System wide vulnerability reports within one-click

Administrative interfaces

  • Cloud Pak for Data web client: depending on user role, you can access either the standalone Cloud Pak for Data web client, or, at a different URL, the Cloud Pak for Data System web client for system administrators, with integrated Cloud Pak for Data web client.
  • System command line interface apcli
For more information, see Administration interfaces.