Super Computing Clusters - Instances| Alibaba Cloud Documentation Center

Super Computing Clusters

Edit in Github

Last Updated: Apr 09, 2019 Edit in Github

This topic describes the features of two Super Computing Cluster (SCC) instance type families: scch5 and sccg5. The instance type family scch5 features high clock speed while sccg5 is for general use scenarios. Their specific instance types are also provided.

Overview

SCCs are based on ECS Bare Metal (EBM) Instances. Given the high-speed interconnects of Remote Direct Memory Access (RDMA) technology, SCCs greatly improve network performance and the acceleration ratio of large-scale clusters. Therefore, SCCs have all the benefits of EBM Instances and offer high-quality network performance featuring high bandwidth and low latency.

SCCs are intended for high performance computing. In the clusters, nodes are connected by RDMA networks featuring high bandwidth and low latency, which guarantee the parallel efficiency demanded by applications that require high-performance computing. The RDMA over Convergent Ethernet (RoCE) rivals an Infiniband network in terms of connection speed and supports more extensive Ethernet-based applications. The combination of SCCs and other Alibaba Cloud computing products such as ECS and GPU instances provides the Alibaba Cloud elastic high-performance computing (E-HPC) platform with the ultimate high performance parallel computing resources, making supercomputing on the cloud a reality.

SCCs are released by Alibaba Cloud to meet the demands of applications such as high performance computing, artificial intelligence, machine learning, scientific or engineering computing, data analysis, and audio and video processing.

Comparison of SCCs to physical and virtual servers

The following table compares SCC, physical servers, and virtual servers.


Metric	Feature	SCC	Physical server	Virtual server
Automated O&M	Delivery in minutes	Yes	No	Yes
Computing	Zero performance loss	Yes	Yes	No
	Zero feature loss	Yes	Yes	No
	Zero resource competition	Yes	Yes	No
Storage	Fully compatible with ECS cloud disks	Yes	No	Yes
	Start from cloud disks (system disks)	Yes	No	Yes
	System disk can be quickly reset	Yes	No	Yes
	Uses ECS images	Yes	No	Yes
	Supports cold migration between physical and virtual servers	Yes	No	Yes
	Requires no installation of operating system	Yes	No	Yes
	Discards local RAID, and provides stronger protection of data on cloud disks	Yes	No	Yes
Network	Fully compatible with ECS VPCs	Yes	No	Yes
	Fully compatible with the ECS classic network	Yes	No	Yes
	Free of bottlenecks for communications between physical and virtual server clusters in a VPC	Yes	No	Yes
Management	Fully compatible with the existing ECS management system	Yes	No	Yes
	Consistent user experience with virtual servers regarding VNC and other features	Yes	No	Yes
	Guaranteed out of band network security	Yes	No	No

scch5, Super Computing Cluster (SCC) instance type family with high clock speed

Features

I/O-optimized
Supports SSD Cloud Disks and Ultra Disks
Supports both RoCE and VPC networks, of which RoCE is dedicated to RDMA communication
With all features of ECS Bare Metal Instance
3.1 GHz Intel Xeon Gold 6149 (Skylake) processors
Equipped with a vCPU to memory ratio of 1:3
Suitable for the following scenarios:
- Large-scale machine learning applications
- Large-scale high-performance scientific and engineering applications
- Large-scale data analysis, batch computing, video encoding

Instance types


Instance type	vCPU	Memory (GiB)	GPU	Bandwidth (Gbit/s)**	Packet forwarding rate (Thousand pps)***	RoCE (Inbound/Outbound) (Gbit/s)	NIC queues****	ENIs*****
ecs.scch5.16xlarge	64	192.0	N/A	10.0	4,500	46	8	32

Note For more information about SCC, see ECS Bare Metal Instance and Super Computing Clusters.

See other instance type families.

sccg5, geneneral-purpose Super Computing Cluster (SCC) instance type family