Application Centric Management of Infrastructure Resources

Cloud Application Management

Subscribe to Cloud Application Management: eMailAlertsEmail Alerts newslettersWeekly Newsletters
Get Cloud Application Management: homepageHomepage mobileMobile rssRSS facebookFacebook twitterTwitter linkedinLinkedIn

Cloud App Mgmt. Authors: Nick Basinger, Elizabeth White, Stefan Bernbo, Yeshim Deniz, Harry Trott

Related Topics: Cloud Computing, Microservices Journal, Cloud Application Management


Fabric Computing and the Cloud

Multi Tenancy - Shared Nothing Pattern and Fabric Computing

Fabric Computing
Fabric computing is defined as a set of computer storage and components joined through a fabric interconnect system.

A fabric computing architecture provides reconfigurable pools of computing, communications and storage resources connected to a highly available, high performance, scalable communications fabric. The fabric computing  architecture provides pools of standard processors, memory, networking, storage and IO which can be assembled  into logical building blocks including VLAN networking, SAN networking, IDS, load balancing, firewall, VPN, router, IO gateways and servers. These building blocks can be assembled in real time to deliver the complete infrastructure to support any application.

Multi Tenancy Shared Nothing Pattern & Fabric Computing
Two of the major attributes of Cloud Computing are:

  • Dynamic Computing Infrastructure: Underlying software and hardware can respond dynamically to changing levels of demand
  • Multi Tenancy: Ability to serve Compute and Storage resources for multiple tenants on demand basis with the tenants abstracted from the physical locations of their resources

With these attributes of Cloud Computing, high matches with that of  Fabric Computing,  the future  data centers that will host cloud based services and ideally  to adapt the  Fabric Computing.

One of the Design Patterns of the Multi tenancy is Shared Nothing architecture. In the Shared Nothing model, resources are divided multiple tenants and each tenant is given the control of its own data, memory locations and storage resources.

The shared nothing model greatly simplifies things  like resource contention, including memory , locks, and processors. Implemented properly  it offers unlimited scalability,  as  new tenants are added or work load increases automatically  resources will be added to the Fabric.

Shared Nothing Multi Tenancy Architecture

Fabric Computing Support

Multi tenancy expects to avoid upfront costs for the tenants, rather  will   have the cost of computing dynamically proportional to the usage. Shared  nothing Architecture facilitates  computing costs than in shared every thing model.

Compute and Storage resources are  granular. Rather than plan for high capital expenditure costs related to individual projects, enterprises will be able to grow workload provision more incrementally.

In Shared Nothing Architecture   tenants are segregated and  failure or specific issues of one tenant  will not affect others  and easier to perform tenant specific  infrastructure activities.

Hardware failure will usually affect only a small part of the total fabric, and user isolation and replacement of failed components will be easier to perform.

Dynamic Scaling is a key component of Cloud Computing and the Shared Nothing architecture facilitates unlimited scalability

A fabric will support dynamic workloads more effectively because resources can be assigned (and reassigned) as workloads demand, which will be increasingly under the control of the management software suite.

Partitioning Schemes & Data Aware Networks (Links)
The success factor of  adopting  Fabric Computing towards the dynamic scalability needs of multi tenant enabled  cloud applications are  only possible if there are robust partitioning algorithms that  allocate  fabric resources  like standard processors, memory, networking, storage and IO are  allocated to multiple tenants in dynamic and transparent way.

Currently without Fabric Computing we find the existing ‘Shared Nothing Architectures' are good at  separating the storage resources among multiple tenants, but currently no easy way to partition other resources like  processors, memory and networking.

Fabric Computing can get there only if the Networks that form the basic of building together a Fabric are more intelligent data-aware devices rather than merely used for data transfer.

Making the networks and routers as part of Fabric more intelligent  will enable the  resource requests for Compute, Processor, Storage are routed appropriately and   the  entire work load is dynamically scalable.  This also allows adding new Fabric Components as New Tenants and Work Loads are added .

Such a partitioning Scheme that will utilize a Fabric is not currently available, but  that should not be something very difficult to implement.

While the Cloud Platform and Multi Tenancy enables transparent access to Shared Pool of resources, so far implementation of the Cloud Platform in the back ground is quite difficult  - base platform, virtualization layer, provisioning layer all are  implemented   using the  existing traditional data center products and level of granularity    required was missing.

Fabric Computing provides the much needed granularity  for implementing the Shared Nothing Architectures for Cloud-based delivery model for multiple tenants. But  it requires stronger management and  stronger algorithms are adding and removing components on demand at a granular level.

More Stories By Srinivasan Sundara Rajan

Highly passionate about utilizing Digital Technologies to enable next generation enterprise. Believes in enterprise transformation through the Natives (Cloud Native & Mobile Native).