Information Technology Reference
In-Depth Information
sluggishness and unavailability without a real concrete way of explaining what is wrong.
The servers are performing at their peak, the network shows minimal congestion, stor-
age systems all show green across the board, so there must be something else wrong.
The only explanation is that it could be the WAN or external Internet end points at the
user's end.
Turning our attention back to things that are under our control, what steps should we
take to ensure network availability and bandwidth on our end? And between what ends
(from the user's end or the provider's end) should we guarantee network and bandwidth
performance? There are no standards for this at this time, but some cloud computing advo-
cates are looking into these questions, so for now, to each his own. Ultimately, the decision
for providing bandwidth performance lies with the provider, and the provider may guaran-
tee this performance in and around only their own jurisdiction, which is to say their own
networks and those in between data centers and availability zones.
Bandwidth and network performance should be a joint venture by different network
providers because it cannot really be guaranteed by one single entity.
Throughput: Bandwidth Aggregation
Bandwidth is an important aspect of a network, especially one made for cloud comput-
ing applications. Bandwidth is directly related to network throughput, which is the rate
of successful message deliveries over a given communications channel. But a single net-
work connection has its limits. Take Ethernet, for example; its bandwidth increases by
10 orders magnitude every generation, so we get 10 Mbit/s, 100 Mbit/s, 1,000 Mbit/s,
and 10,000 Mbit/s. When you reach the bandwidth ceiling of one generation, you simply
move to the next generation for increased throughput. However, doing so can be cost
prohibitive, especially for the latter generations. So if your need exceeds the limit for
1,000 Mbps, you can move your infrastructure to 10,000 Mbps, which would then cost
a lot because you have to change a lot of elements in your infrastructure. A good way to
circumvent this cost is by teaming or bonding, more officially known as link aggregation.
Bonding
Bonding, or NIC bonding, is a form of bandwidth aggregation where multiple smaller
connections act as a single big connection. This requires multiple NICs working together
in a single machine. This is applicable for personal computers, but servers, having multiple
NIC slots, can make better use of bonding.
For bonding to work, there must be a server that breaks up data into multiple packets
to send to a client. If there is a bonded connection between them, the packets are simulta-
neously sent over these different connections and then reassembled on the receiving end,
which means they arrive more or less at the same time, or at least it takes less time for them
to arrive through multiple connections as opposed to a single connection where they arrive
almost in a serialized manner, even though in essence it is not actually serialized. That is
the reason it takes longer to download large files; your machine will be waiting for all the
packets to arrive and there is only so much bandwidth, so they have to come through one
Search WWH ::




Custom Search