IDH on Cisco UCS Common Platform Architecture
Raghu Sakleshpur is an engineering manager at Intel who works on Hadoop deployments and Big data technologies with partners, ISVs and customers. He is a technologist to the core and loves to share his experiences on Big data and Hadoop technologies whenever the opportunity presents itself. In his spare time, he loves pursuing his other passions like running, hiking, biking and traveling.
Intel has partnered with Cisco to announce a comprehensive big data analytics platform that is performance driven and scales for mission-critical applications using the Intel distribution of Hadoop (IDH) running on Intel Xeon processor powered hardware of Ciscoâ€™s Unified Computing Systemâ€™s (UCS) common platform architecture.
This best of the breed mix of technologies from both companies helps reduce the time and risk associated with Hadoop deployments and with enhanced features, provides an optimized solution for outstanding performance and Scalability.
Below the Intel distribution of Hadoop stack, the cluster running on Cisco common platform architecture of the UCS line of cluster racks is made up of the following components:
- Cisco UCS C240 M3 Rack-Mount Servers: Each Hadoop node in the cluster will be a 2-RU server, powered by two Intel Xeon E5-2600 series processors that can support up to 768 GB of main memory and include up to 24 small form factor disk drives for performance optimized configurations or include 12 large form factor drives for capacity optimized configurations. They also feature Cisco UCS VNIC optimized for high bandwidth and low latency cluster connectivity with support for up to 256 virtual devices.
- Cisco UCS 6200 Series Fabric Interconnects: This provides high speed low latency connectivity for all nodes in the cluster and centralized manager of all nodes of the cluster via the UCS manager.
- Cisco UCS 2200 Series Fabric Extenders which act as remote line cards for fabric interconnects that provides a cost efficient scalable connection for hundreds of nodes in the Hadoop cluster.
- Cisco UCS Manager which provides unified embedded management of all components in the Cisco UCS common platform architecture.
IDH on UCS is offered in single rack or multiple rack configurations depending upon the scale and size of the Hadoop cluster. The single rack configuration consists of two Cisco UCS 6248UP 48-Port Fabric Interconnects that can support up to five racks or two Cisco UCS 6296UP 96-Port Fabric Interconnects that can support up to ten racks. Each server in the cluster connects to the unified fabric through two active-active 10-GigE links using a Cisco UCS VNIC. Multi rack configurations include two Cisco Nexus 2232PP fabric extenders and 16 Cisco UCS C240 M3 Rack-Mount servers for every additional rack.
A typical IDH cluster on UCS Common platform architecture that is performance optimized can support up to 256 Intel Xeon cores and 32 Gigabytes/sec (SATA) or 48 Gigabytes/sec (SAS) I/O bandwidth. Whereas a typical capacity optimized IDH cluster can support up to 576 TB of raw storage per rack and about 5.76 petabytes per UCS domain that can span up to ten racks.
Scaling the IDH cluster beyond 10 racks (beyond 160 servers) is achieved by interconnecting multiple UCS domains using the Cisco Nexus 6000/7000 Series switches. Such IDH clusters can scale to thousands of nodes with hundreds of petabytes of storage.
More discussion on IDH integration with Cisco UCS CPA and the Cisco Validation document that provides a blue print to build Hadoop clusters will be covered in the next blog.