7
Cloud Networking for Big Data Processing How networking works in massive scale of cloud networks? Datacenter 10’s and thousands of servers connecting into a networking, consolidating compute into such large facilities use benefits from multiplexing. You are sharing the infrastructure for the many applications, it also uses the economy of scale Networking in the Cloud? Cloud computing brings new challenges for the Cloud Networking These networks are huge, 10’s of thousands even 100 thousand servers in the largest datacenters today And applications these machines are running are big data processing applications need much more high performance connectivity between the servers not just occasional communication with the client out to the internet than in traditional datacenters So both the challenges scale of the network as well as demand of the applications placed on them means that traditional techniques for networking clusters just don’t work in these settings

Cloud Network

Embed Size (px)

Citation preview

Page 1: Cloud Network

Cloud Networking for Big Data ProcessingHow networking works in massive scale of cloud networks?

Datacenter 10’s and thousands of servers connecting into a networking, consolidating compute into such large facilities use benefits from multiplexing. You are sharing the infrastructure for the many applications, it also uses the economy of scale

Networking in the Cloud?

Cloud computing brings new challenges for the Cloud Networking

These networks are huge, 10’s of thousands even 100 thousand servers in the largest datacenters today

And applications these machines are running are big data processing applications need much more high performance connectivity between the servers not just occasional communication with the client out to the internet than in traditional datacenters

So both the challenges scale of the network as well as demand of the applications placed on them means that traditional techniques for networking clusters just don’t work in these settings

Design for the largest datacenters today and see how they address these challenges.

Will see both how switches and networks are connected together into a physical network and also how we deal with that traffic through that network.

Page 2: Cloud Network

Networks need to handle traffic varies across different times, further one such large collection of devices from infrastructure failures occurs often they are not just common case, they are way of life.

So these concerns call for efficient and fault tolerant traffic engineering.

On top of such infrastructure run variety of applications including those for big science but also pretty much every web application you can think of

Page 3: Cloud Network

Running these application require big data processing across machines behind the scenes but also they have new kinds of real demands driven by interaction with us humans, so whenever you visit a website it appears to be one interaction from your perspective but behind the scenes in the cloud infrastructure there can be hundreds or even thousands of queries running at the backend server in order to assemble response and give you the web page you have asked for.

So we need to examine the traffic pattern the application creates and challenges that brings the networking in these settings. The nature of these traffic patterns is another key difficulty in cloud networking.

To give you one example, we need to learn and design how TCP can be designed to push to its limits by traffic characteristics and datacenters. And how the resulting performance problems are dealt with today in new ways.

Isolation and Security in Cloud Network

Given all this infrastructure to be shared not only in the application but across the network and for paying cloud users we need isolation across the network not only in terms of performance but also for security. This is another major which form cloud networking.

We can address this need through networking virtualization, further more with large number of devices and dynamic environment of datacenters old techniques are not going to work.

Today we can address them with software defined networks.

Page 4: Cloud Network

Architecting Software Defined Networking

Software defined networking exposes the abstract centralize view of network on top of which networking management functions become programmable

Separation between humans and Data is another Challenge for the cloud network.

Page 5: Cloud Network

To deal with these largest datacenter service providers run multiple datacenters across the globe to provide services to the clients in different geographies. Serving clients from Sydney to New York otherwise would not be efficient.

Cutting Edge Traffic Engineering technique allows you to make efficient use of this expensive bandwidth. Even then few datacenters across the globe doesn’t help to get close to the users in order to achieve that you may need thousands of Datacenters and CDN.

My Research & Work continues …

Asif, Muhammad (International Technical Solution Expert [email protected] )