To meet increasing customer demands of cloud scale, Microsoft is deploying hundreds of thousands of new servers every year, and the company has spent more than $15B building datacenters around the world. At the center of this virtuous cycle is incredible innovation in software, hardware, and processes to address costs of procuring, deploying, operating, and retiring hardware at this scale.
Since 2014, Microsoft has contributed these designs and open source tools to OCP {e.g., server (OCS); back up power with LES (Local Energy Source), etc.}. Microsoft continues to work with System Providers and Component Suppliers to address cost, easy of deployment, and continuity of supply challenges. During the presentation Sanjeev Khanna will talk about how Microsoft is working is broaden the ecosystem and how OCP community can take advantage of these innovations and our learnings.
Traditional monolithic storage doesn’t align with today’s open source scale-out databases and big data frameworks. But creating scalable open-compute clusters with the latest storage building blocks demands expertise in balancing capacity, I/O, throughput, latency and quality of service against budgetary constraints.
In the early days of open computing, simple concepts like 1:1 core:spindle ratios made it easy to design commodity-based infrastructures. But in our quest to build vanity free servers, well-intentioned software functions like MySQL/NoSQL sharding and Hadoop replication have compounded server/storage sprawl, driving up space, power and cooling (OpEx). This unintended consequence has restricted our CapEx budgets for new equipment needed to tackle the next big thing.
With exponential improvements in SSD price/performance and relentless efforts to expand HDD capacities and reliability, 1:1 is no longer the right metric for open compute designs.
This presentation will provide an unbiased view of hot, cool and cold data storage trends, along with benchmarking approaches and new metrics for addressing budget realities. We will then share construction blueprints for open compute solutions using the latest in SSD, HDD and software technologies that optimize efficiency and fidelity at scale.
There are several major rackscale open hardware platforms currently available, including OCP (Facebook Implementation), OCS (Microsoft Implementation), Scorpio Project Designs (Open Datacenter Designs for China), Intel RSA (Rack Scale Architecture). Datacenter and Cloud designers need to be aware of the differences among them and choose the one which is most suitable for their use case to deploy various types of private cloud, public cloud and hybrid cloud. We will invite Baidu as case study to speak about their datacenter deployments based on Scorpio Project Designs.
http://www.youtube.com/watch?v=aRnh_e3wpDc
Telco and data centre operators want software-defined capabilities to match their newly open hardware. This talk will demonstrate the state of the art for top-of-rack switches, as well as forthcoming telco-centric open networking such as DSLAMs, GPONs and software-defined LTE base stations, that build a software ecosystem to take advantage of OCP’s open hardware strategy.
Using open source mechanisms optimised for scale-out environments, the speakers will walk through a model central office or data centre, showing how diversity of NOS options and diversity of networking apps can fit together in a highly-automated cluster. Using OpenBMC and FBOSS from Facebook, MAAS, LXD containers, Ubuntu and a range of apps, the speakers will demonstrate dramatically more efficient software-defined approach to infrastructure.
We will use OCP switches to demonstrate running apps on the ToR switch that deploy and manage both server and physical switch infrastructure. Using OpenBMC, PXE, and ONIE, we are able to automate the deployment of scale-out bare metal infrastructure using 100% open source tools. We will show how the separation of OS and apps enables greater competition for network control plane vendors as well as creating new opportunities for innovation in the SDN, NFV and network intelligence markets.