Practical Cloud Knowledge: Install, Configure, and Manage - Deploying and Managing a Cloud Infrastructure - page 210

Cloud Virtual Router The cloud virtual router is the logical counterpart of a physical
router in the same vein as virtual machines, virtual NICs, and virtual switches. It serves the
same purpose in the virtual network as that of the physical one in the physical network. It
is used for routing, but its main draw is easier administration of security such as firewalls
and of network policies and using a familiar technology. But when we say virtual router,
it could mean different things to different people because the concept has been around for
a time. Even before proliferation of cloud computing, it has been used to mean a software
implementation for allowing a host computer to act as a router through its regular NIC or
wireless NIC.
The term virtual router still mostly refers to the software implementation described earlier;
that is why we may specify that we mean cloud virtual routers, those that exist in hypervisors
and used in virtualized environments to connect multiple virtual machines together. Virtual
routers also serve to lessen the clutter that making a single host run hundreds of virtual
machines creates. When there are so many virtual machines and virtual networks present, a
routing virtual switch often is not up to the task because it really is not meant for that job.
The switch keeps tabs on the MAC address of each NIC connected on its ports and is usually
suitable for local networks with devices directly connected to it, but a router not only routes
packets, it also ensures that they reach their destination using the shortest path possible and
is responsible for connecting different networks together because of this capability. So when
we have multiple networks within the virtual environment, the only logical way of connecting
them to each other and to the outside network is through virtual routers that act as gateways
between each virtual network.
Shared Memory
Cloud computing architecture demands that multiple machines and multiple availability
zones can act together as one, able to consolidate resources into a large pool. Storage area
networks, or SANs, do this for disk storage, but there is currently no SAN equivalent for
main memory. That means that memory still resides in each individual physical machine or
server, so there is a need for all of these machines to be able to pull their collective memory
capacity into a pool and then be able to reallocate it to individual virtual machines that
need it. This is quite important because a virtual machine can exist even when its individ-
ual pieces are scattered in different machines geographically. This is one way of implement-
ing shared memory, specifically known as distributed memory, for redistribution, but it is
not yet implemented, so currently virtual machines reside in a single host machine where
they get their processing power (virtual CPU) and memory (virtual RAM), while the file
that makes up the virtual hard disk can be striped across multiple machines or even SANs.
Although, that would cause a bit of latency depending on the system, but it can and has
been done.
A shared and distributed memory view does not distinguish from local or remote data,
which can cause performance degradation, especially if the memory is allocated from geo-
graphically separated nodes. This actually is counterintuitive to the goal of main memory,
which is to be a quick and physically near or close to the CPU (module). But the concept is