Here is, verbatim, an article I posted on Linked-In yesterday [For other posts on Linked-In, view my Linked-In profile] :
Imagine an 8-lane highway. Now imagine a 4-lane highway. Which has the greater bandwidth ?
Imagine your organisation sends its employees on a wekend “retreat” by bus. You have the choice of two locations, one that is 200kilometres away and the other is 80kilometres away. Assume that buses travel at a constant speed of 80kmph. Which resort will your employees get to faster ?
The first question is about bandwidth. The second is about latency.
(Why should I assume a fixed speed for the buses ? Because, I can assume a fixed speed at which electrons transfer over a wire or photons over a light channel).
Expand the question further. What if the organisation needs to send 32 employees in a 40-seater bus. Does it matter that the bus can travel on an 8-lane highway versus a 4-lane highway (assuming minimal other traffic on the highways at that time) ?
Too often, naive “architects” do not differentiate between the two. If my organisation needs to configure a standby (DR) location for the key databases and has a choice of two locations but varying types of network services, it should consider *both* bandwidth and latency. If the volume of redo is 1000MBytes per minute and this, factoring overheads for packetizing the “data”, translates to 167Mbits per second, should I just go ahead and buy bandwidth of 200Mbits per second ? If the two sites have two different network services providers offering different bandwidths, should I simply locate at the site with the greater bandwidth ? What if the time it takes to synchronously write my data to site “A” is 4ms and the time to site “B” is 8ms ? Should I not factor the latency ? (I am assuming that the “write to disk” speed of hardware at either site is the same — i.e. the hardware is the same). I can then add the complications of network routers and switches that add to the latency. Software configurations, flow-control mechanisms, QoS definitions and hardware configuration can also impact bandwidth and latency in different ways.
Now, extend this to data transfers (“output” or “results”) from a database server to an application server or end-user. If the existing link is 100Mbps and is upgraded to 1Gbps, the time to “report” 100 rows is unlikely to change as this time is a function of the latency. However, if the number of concurrent users grows from 10 to 500, the bandwidth requirement may increase and yet each user may still have the same “wait” time to see his results (assuming that there are no server hardware constraints returning results for 500 users).
On the flip side, consider ETL servers loading data into a database. Latency is as important as bandwidth. An ETL scheme that does “row-by-row” loads relies on latency, not bandwidth. Increasing bandwidth doesn’t help such a scheme.
Think about the two.