Storage Provisioning and Networking - Deploying and Managing a Cloud Infrastructure - page 260

Typically, enterprises lease common carrier circuits to connect MANs instead of laying
out their own cables (which can be expensive). MANs can also be connected via wireless
technologies such as radio or infrared (IR).
Network Operations
Network operations involve handling bandwidth requirements and limiting latency as well
as handling compression, caching, and load balancing. Generally, companies assign band-
width quotas to user groups. To load balance Internet usage, multiple Internet connectivity
providers are contracted. For instance, a large enterprise might have a high-capacity link
for its production systems and a general usage network link for office usage.
Latency requirements and caching levels are defined on per-application or -system basis.
This is mostly done by keeping in perspective the usage expected for each production-level
application or system, such as the amount of clients (incoming), amount of queries (incom-
ing), amount of responses (outgoing), expected response time per query, typical user opera-
tions (e.g., 8 of 10 are read), nature of operations (read or write), most used application,
least used application, and so on.
Cloud Storage Technology
Cloud storage is a model of networked storage for enterprise-level systems. Data is usually
distributed over a large number of virtualized storage resources. A typical data center would
host large chunks of data from various sources. This is not true in all cases, but very large
quantities of data (usually hundreds of petabytes) might physically span not only across
multiple servers in the same data center but also across multiple geographical locations.
In order for their customers to use cloud storage services for accessing data, companies
operating in the data center marketplace offer web services, content management systems
(CMSs), APIs, desktop software, and cloud storage gateways (cloud storage gateways are
discussed later in this chapter).
Cloud storage provides all the cloud features such as scalability, agility, elasticity, and
multitenancy. Perhaps the greatest feature is the federated data storage, where multiple
virtualized data storage resources appear and behave as a single unit. This is further sup-
ported by fault tolerance and durability through redundant replicas.
However, redundancy does come at a cost. Most of the data storage systems use the
primary storage location to read and write data transactions. The data replicas are only
guaranteed to be eventually consistent. I/O operations (including cross-network) are still
expensive, and pushing each singleton update would overload the data and network buf-
fers throughout the route, thus rendering the system useless. This might even cause a chain
reaction and cause disruption of services throughout the platform. I/O write updates (data
manipulation) are carried to replicas in batches of designated sizes to minimize network
costs and prevent possible service disruption.