Ashwin's Roam Garden
Powered by š±Roam GardenScalability and Convergence of Data Center IT hardware
1. Introduction
Hierarchical representation of a data center
Power, IT, Cooling
How we build ācomputersā or computing machines
Warehouse-scale computers (WSCs)
name is meant to call attention to the most distinguishing feature of these machines: massive scale of their software infrastructure, data repositories, and hardware platform (Barroso, 2019)
Micro/Modular Data Centers (MDC)
Service level agreement (SLA) leading to cluster of racks within a data hall
Software-defined data centers (SDDC) (Levy et. al, 2019) Choice of computing machine boundaries is application-driven and increases interconnection and interdependence between various levels of the data center
Anomalous workloads, equipment upgrades, data center layout changes
The evolving form factors: micro-modular, modular
Dell EMC
2. Overview
2.1 Problem Statement
Deployment of IT modules is largely a function of the end-userās application requirement. An end-user can turn to an Original Equipment Manufacturer (OEM) to source the necessary IT equipment and hardware. The range of products OEM companies offer will reflect the needs of such potential end-users. The OEMās ensure their IT hardware is offered in forms that are customer-centric, scalable and easily deployable.
Understanding the computing machine boundaries that emerge out of this end-user and OEM environment can provide crucial guidance in tailoring energy-efficient, but disruptive, cooling technologies such as direct or indirect evaporative cooling (D/IEC). Aside from the large-scale proprietary cooling solutions available, the development effort of D/IEC is mainly driven by large enterprises who use relatively homogenous IT hardware and system software platforms and share a common systems management layer.
However, the conventional data centers are hindered by the lack of deployment flexibility, predominance of third-party software, Service Level Agreements (SLA) and the overarching uncertainty in incorporating energy efficient cooling solutions.
This project aims to conduct a scaling analysis on how these computing machine boundaries are being drawn in the modular data center sector. Module, cabinet/rack and equipment level form factors will be considered. Factors such as Edge computing and 5G driving the emergence of micro-modular and modular data centers will be surveyed. A hierarchical (top-down and bottom-up) scaling analysis will be carried out at module, cabinet/rack and equipment levels to study the emergent form factors and how they scale with end-user application/service and the type of cooling technology adopted.
3. Project Scope
3.1Ā
4. Project Goals and Objectives
What is the scaling parameterā?
Identify the critical application-driven factors that influence how a system (at different hierarchical levels) is proportioned with application-specific resources (compute, storage, flash, graphics etc) as it scales the following:Ā
Power, cooling, ācomputeāĀ
Latency
What are the enabling cooling technologies? What are the disruptive cooling technologies?
{{table}}
Level of Analysis
Inputs
Tools & Techniques
Outputs
Equipment
Cabinet/Rack
Modules
Data Hall
Level of Analysis (LOA) and model requirements
Model for understanding, Model for prediction
Questions posed determine complexity (measured via computational, experimental, level of prior knowledge about the system)
4.1 LOA : EquipmentĀ
Develop a server compact model that is sensitive to the level of detail required in studying the system behavior
Survey to determine form factor and power proportionality trends in ITE
For the survey, the first step would be to search, review and record all the readily available characteristics and properties of servers. Currently, the plan is to be able to simplify the server representation in our CFD models with the emphasis on what system level response (real behavior based on control or failure sequence) we intend to mimic/capture in our model.
Here are some of the questions we will be able to answer upon completing this survey:
Questions When do you switch/upgrade from one cooling technology to another? (Ex. Air to liquid cooling.)
The answer could depend on what utilization ratio of compute vs storage/memory you are looking for and at what form factor constraint.
Questions What are the key design parameters for extending air coolingās limitations?
4.1.1 VirtualizationĀ
4.2 LOA : Cabinet/Rack
4.3 LOA: ModulesĀ
Understand the end-user needs to āModularizeā
Rack deployment scheme
Number of total racks per modular deployment
Corresponding rack density increments