At first it might seem like latency and bandwidth are directly related to cost, for example that high latency always means a high cost, or a low bandwidth always means a high cost. However, this is not always correct.
One example where there can be high latency, but low cost, is if we employ latency/stall hiding as described in lecture 2. On slide 47 in particular, it shows an example where one core is given 4 threads, and every time a thread hits a stall because it has to go out to memory (which has high latency), it switches context to a different thread. The total time for the operations is slightly longer than if there had been no stalls at all (which reflects that going to memory did have some cost), but it is significantly less than if there had been no context-switching. This shows that although the latency for each memory access was high, the total cost for all of them was relatively low.
This comment was marked helpful 0 times.
paraU
This greatly depends on your definition of "cost". If you define it as time spent on finishing all the jobs, then the cost is smaller because throughput is increased. However, if you define your cost as time spent on finishing this specific job, your cost has actually increased because of the thread scheduling and context switching. So it really depends on the definition of "cost".
This comment was marked helpful 0 times.
elemental03
At first i was confused by the terms bandwidth and latency because i was more used to talking about them in terms of network latency and bandwidth, not in terms of a singular operation or program. So i did some research on the network latency and bandwidth and figured i would share what i found in a laymen's terms for those who might be mixed up like i was (the below notes are taken from various sources):
What is network latency?
It describes the delays involved with Internet connectivity. A lower latency simply means a lower delay time, giving us the perception of faster network speed, although it might not actually be faster technically. A high-latency connection has more delays and performs more slowly. You can measure latency with ping tests to determine how many milliseconds it takes for a test data set to travel from one computer to another. Gamers know what i'm talking about
So why is it important?
High-latency levels create data bottlenecks and unacceptable lag that hurts the performance of the entire network. High-latency networks in an office, for example, can hurt productivity and hinder your ability to respond in an environment that demands fast response times, such as the news media or the stock exchange (I think this example was brought up in lecture too). Gamers always try to modify their computers to chase lower latency and are constantly modifying their connections to reduce lag and enhance their performance in their favorite games.
What is bandwidth?
Lets look at your internet connection as an actual pipeline. Then, bandwidth would be the width and breadth of your pipe. A larger pipe means you can send more data though. Therefore, more bandwidth means that you can pass a lot more data through your connection in the same amount of time. Most Internet service providers represent bandwidth with a measurement of megabytes per second.
Why is bandwidth important?
The more bandwidth you have, the more information you can send through the Internet without realizing a slowdown or reduced performance. For example, if you're on a 30 Mbps connection, you can tax it with multiple downloads and traffic and realize excellent performance as long as you don't exceed a combined 30 Mbps limit.
At first it might seem like latency and bandwidth are directly related to cost, for example that high latency always means a high cost, or a low bandwidth always means a high cost. However, this is not always correct.
One example where there can be high latency, but low cost, is if we employ latency/stall hiding as described in lecture 2. On slide 47 in particular, it shows an example where one core is given 4 threads, and every time a thread hits a stall because it has to go out to memory (which has high latency), it switches context to a different thread. The total time for the operations is slightly longer than if there had been no stalls at all (which reflects that going to memory did have some cost), but it is significantly less than if there had been no context-switching. This shows that although the latency for each memory access was high, the total cost for all of them was relatively low.
This comment was marked helpful 0 times.
This greatly depends on your definition of "cost". If you define it as time spent on finishing all the jobs, then the cost is smaller because throughput is increased. However, if you define your cost as time spent on finishing this specific job, your cost has actually increased because of the thread scheduling and context switching. So it really depends on the definition of "cost".
This comment was marked helpful 0 times.
At first i was confused by the terms bandwidth and latency because i was more used to talking about them in terms of network latency and bandwidth, not in terms of a singular operation or program. So i did some research on the network latency and bandwidth and figured i would share what i found in a laymen's terms for those who might be mixed up like i was (the below notes are taken from various sources):
What is network latency? It describes the delays involved with Internet connectivity. A lower latency simply means a lower delay time, giving us the perception of faster network speed, although it might not actually be faster technically. A high-latency connection has more delays and performs more slowly. You can measure latency with ping tests to determine how many milliseconds it takes for a test data set to travel from one computer to another. Gamers know what i'm talking about
So why is it important? High-latency levels create data bottlenecks and unacceptable lag that hurts the performance of the entire network. High-latency networks in an office, for example, can hurt productivity and hinder your ability to respond in an environment that demands fast response times, such as the news media or the stock exchange (I think this example was brought up in lecture too). Gamers always try to modify their computers to chase lower latency and are constantly modifying their connections to reduce lag and enhance their performance in their favorite games.
What is bandwidth? Lets look at your internet connection as an actual pipeline. Then, bandwidth would be the width and breadth of your pipe. A larger pipe means you can send more data though. Therefore, more bandwidth means that you can pass a lot more data through your connection in the same amount of time. Most Internet service providers represent bandwidth with a measurement of megabytes per second.
Why is bandwidth important? The more bandwidth you have, the more information you can send through the Internet without realizing a slowdown or reduced performance. For example, if you're on a 30 Mbps connection, you can tax it with multiple downloads and traffic and realize excellent performance as long as you don't exceed a combined 30 Mbps limit.
This comment was marked helpful 0 times.