Notes for Infrastructure Investments blog post:
This is a list of points from my notebook over the past year collected over various meetings and conferences.
Mobility
- Consumer mobile behaviour matters – leading indicator broadly for mobility
- Think beyond current devices (3-5″ smartphones, 7-11″ tablets) – what devices/screens matter in 2015 ?
- Pay attention to the rise of infrastructure apps (How mobility connects information silos)
- Composite Apps matter more vs. individual apps (IFTTT + hardware + ambience awareness)
- Role of data in mobile architectures (Virtual cell definitions, moving beyond ‘circuit’ connections to a single base station). Multiple radios (WiFi, 4G/LTE, …)
- Offloading mobile traffic to data-centers vs. core-networks.
- Mobile is not a “second screen”
- Think “Interaction” when you think of mobile screens, not “presentation”
Cloud & Control
- Management & Control frameworks for heterogenous hardware
- Service-to-service information exchange with policy/compliance
- Stupid simple ways to deliver app-aware performance (no QoS please), solve by sufficiency/availability of resources, not strict reservation.
- Cloud-to-cloud resource signaling/advertisement/reservation
- Software defined networking v1.0 was MPLS (remember Ipsilon), pay more attention to protocols vs. systems/boxes. Global knowledge neither required, nor assumed for optimal/practical TE.
- Data-center to network boundary+Control matters.
Custom Hardware
- Software defined hardware (is there any other kind?)
- Processor controlled modules for specific workloads (across consumer/enterprise/ServiceProvider/DataCenters)
- Software-defined Networking hardware required: backplanes, Top-of-rack switches, Data-center fabrics, DC-to-DC core networking vs. CO-to-CO (Flows/mobile-traffic/…)
- IO bottlenecks need to be solved – scale (Users/apps) begets throughput problems.
Data Intelligence
- New BI stack on Google/Amazon infrastructure vs. specialized warehouses
- DI stack = presentation/visualization + Infrastructure-smarts (SW, HW) + Federated DI warehouses + DR/HA + flex-scalable db + Caching + …
- Optimize cost per bit/byte of [store, manipulate, move, serve]
- Infrastructure apps play a big role here, as does custom hardware (workload specific compute/store/network)
- 2000-2010 was v0.1 (MapReduce), think Dremel, Cassovary, Spark & Shark,…