Cloud-Based Infrastructure for Data-Intensive e-Science Applications: Requirements and Architecture - Cloud Computing with e-Science Applications - page 32

Layer D4: (Shared) scientific platforms and instruments specific for dif-
ferent research areas.
Layer D5: Access and delivery layer that represents the general Federated
Access and Delivery Infrastructure (FADI) that includes infrastruc-
ture components for interconnecting, integrating, and operating com-
plex scientific infrastructure to support project-oriented collaborative
groups of researchers
Layer D6: Scientific applications, subject-specific databases, and
user portals/clients
Note: The D prefix denotes the relation to the data infrastructure.
We also define the three cross-layer planes: operational support and man-
agement system, security plane, and metadata and life cycle management.
The dynamic character of SDI and its support of distributed multifaceted
communities are guaranteed by the following dedicated layers: D3, the infra-
structure virtualization layer that typically uses modern cloud technologies,
and D5, the FADI layer that incorporates related federated infrastructure
management and access technologies [21, 35, 36]. Introduction of the FADI
layer reflects current practices in building and managing complex SDIs (and
enterprise infrastructures) and allows independently managed infrastruc-
tures to share resources and support interorganizational cooperation.
Network infrastructure is presented as a separate lower layer in e-SDI, but
dedicated network infrastructure provisioning is also relevant to the FADI
layer. Network aspects in big data are becoming even more important than
for computer grids and clouds. We can identify two main challenges that big
data transport will impose on the underlying layer of the SDI:
• Timely delivery to bring all data where required with the smallest
possible latency
• Cost reduction to optimize the amount of network equipment
required (either via purchasing it or on a pay-per-use basis) without
scarifying the quality of service (QoS)
For many SDIs, the basic best-effort Internet is the only available network
transport architecture. In these cases, given the constraints imposed by this
shared medium, it will be difficult to fully provide the low latency and guar-
anteed delivery required for big data processing. Performance may be lower,
but it will be manageable. Fewer SDIs will rely on circuit-based networks, for
which the timely delivery of data will be guaranteed but the costs for operat-
ing or using the network path will be significantly higher.
We see a third possibility for dealing with big data at the lowest layer of the
SDI. Emerging protocols for network programmability (e.g., OpenFlow and
in general software-defined networks) provide interesting solutions. By fully
controlling the network equipment, both time and costs can be optimized.