I often heard people talking about a network's speed in terms of "bandwith", and I read from < Computer Networks: A Systems Approach > the following definiton:
The bandwidth of a network is given by
the number of bits that can be
transmitted over the network in a
certain period of time.
AFAIK, the word "bandwith" is used to describe the the width of frequency that can be passed on some kind of medium. And the above definition describe something more like a throughput. So is it mis-use?
I have been thinking about this question for some time. I don't know where to post it. So forgive me if it is off topic.
Thanks.
Update - 1 - 9:56 AM 1/13/2011
I recall that, if a signal's cycle is smaller in time domain, its frequency belt will be wider in frequency domain, so IF the bit rate (digital bandwidth) is big, the signal's cycle should be quite small, and thus the analog bandwidth it required will be quite wide, but medium has its physical limit, the medium has the widest frequency it allows to pass, so it has the biggest bit rate it allows to transmit. From this point of view, I think the mis-use of bandwidth in digital world is acceptable.
The word bandwidth has more than one definition:
Bandwidth has several related meanings:
Bandwidth (computing) or digital bandwidth: a rate of data transfer, throughput or bit rate, measured in bits per second (bps), by analogy to signal processing bandwidth
Bandwidth (signal processing) or analog bandwidth, frequency bandwidth or radio bandwidth: a measure of the width of a range of frequencies, measured in hertz
...
With both definitions having more bandwidth means that you can send more data.
In computer networking and other digital fields, the term bandwidth often refers to a data rate measured in bits per second, for example network throughput, sometimes denoted network bandwidth, data bandwidth or digital bandwidth. The reason is that according to Hartley's law, the digital data rate limit (or channel capacity) of a physical communication link is proportional to its bandwidth in hertz, sometimes denoted radio frequency (RF) bandwidth, signal bandwidth, frequency bandwidth, spectral bandwidth or analog bandwidth. For bandwidth as a computing term, less ambiguous terms are bit rate, throughput, maximum throughput, goodput or channel capacity.
(Source)
Bandwidth is only one aspect of network speed. Delay is also important.
The term "bandwidth" is not a precise term, it may mean:
the clock frequency multiplied by the no-of-bits-transmitted-in-a-clock-tick - physical bandwidth,
minus bytes used for low-level error corrections, checksums (e.g. FEC in DVB),
minus bytes used by transmit protocol for addressing or other meta info (e.g. IP headers),
minus the time overhead of the handshake/transmit control (see TCP),
minus the time overhead of the administration of connection (e.g. DNS),
minus time spent on authentication (seeking user name on the host side),
minus time spent on receiving and handling the packet (e.g. an FTP server/client writes out the block of data received) - effective bandwidth, or throughput.
The best we can do is to always explain what kind of bandwidth we mean: with or without protocol overhead etc. Also, the users are often interested only in the last brutto value: how long does it take donwloading that stuff?
Related
Hello all I am new to networking and a question arose in my head. Would a device that is physically closer to another device transfer a file quicker than a device which is across the globe if a P2P connection were used?
Thanks!
No, not generally.
The maximum throughput between any two nodes is limited by the slowest interconnect they are using in their path. When acknowledgments are used (eg. with TCP), throughput is also limited by congestion, possible send/acknowledgment window size, round-trip time (RTT) - you cannot transfer more than one full window in each RTT period - and packet loss.
Distance doesn't matter basically. However, for long distance a large number of interconnects is likely used, increasing the chance for a weak link, congestion, or packet loss. Also, RTT inevitably increases, requiring a large send window (TCP window scale option).
Whether the links are wired, copper, fiber, or wireless doesn't matter - the latter means there's some risk for additional packet loss, however. P2P or classic client-server doesn't matter either.
To what extent the server being closer to the user decrease the latency between the user and the service and help the overall experience ?
That depends greatly on what you mean by "closer."
If the two devices are on the same wire, the latency is negligible since the signal travels in copper at about one-third the speed of light. The latency due to distance can be calculated based the distances of the various media in the path. Moving devices closer, from the perspective of cabling distance, depending on the distance, may have a noticeable difference in latency.
Having multiple devices (switches, routers, etc.) in between the two will increase the latency more due to serialization/de-serialization, encapsulation/de-encapsulation, switching, buffering, etc. delays. This is much more complex, and probably has a more significant increase in the latency.
DCTCP is a variant of TCP for Data Center environment. The source is here
DCTCP using ECN feature in commodity switch to limit queue length of buffer in switch around the threshold K. Doing so, packet loss is rarely happen because K is much smaller than buffer's capacity so buffer isn't almost full.
DCTCP achieve low latency for small-flows while maintaining high throughput for big-flow. The reason is when queue length exceeds threshold K, a notification of congestion will be feedback to sender. At sender, a value for probability of congestion is computed over time, so sender will decrease sending rate correspondingly to the extent of congestion.
DCTCP states that small queue length will decrease the latency or the transmission time of flows. I doubted that. Because unless packet loss leading to re-transmission and so high latency. In DCTCP, packet loss rarely happens.
Small queue at switch forces senders to decrease sending rates so force packets to queue in TX buffer of senders.
Bigger queue at switch make senders have higher sending rates and packets instead queue in TX buffer of senders, it now queue in buffer of switch.
So I think that delay in both small and big queue is still the same.
What do you think?
The buffer in the switch does not increase the capacity of the network, it only helps to not loose too much packets if you have a traffic burst. But, TCP can deal with packet loss by sending slower, which is exactly what it needs to do in case the network capacity is reached.
If you continuously run the network at the limit, the queue of the switch will be full or nearly full all the time, so you still loose packets if the queue is full. But, you also increase the latency, because the packet needs some time to get from the end of the queue where it arrived to the beginning where it will be forwarded. This latency again causes the TCP stack to react slower to congestion, which again increases congestion, packet loss etc.
So the ideal switch behaves like a network cable, e.g. does not have any buffer at all.
You might read more about the problems caused by large buffers by searching for "bufferbloat", e.g. http://en.wikipedia.org/wiki/Bufferbloat.
And when in doubt benchmark yourself.
It depends on queue occupancy. DCTCP aims to maintain small queue occupancy, because the authors think that queueing delay is the reason of long latency.
So, it does not matter how maximum size of queue is. In 16Mb of maximum queue size or just 32kb of maximum queue size, if we can maintain queue occupancy always around 8kb or something small size, queueing delay will be the same.
Read a paper, HULL from NSDI 2012, of M. Alizadeh who is the first author of DCTCP. HULL also aims to maintain short queue occupancy.
What they talk about small buffer is, because trends of data center switches shift from 'store and forward' buffer to 'cut-through' buffer. Just google it, and you can find some documents from CISCO or somewhere related webpages.
I found that time used for MPI_scatter/MPI_gather continuously increased (somehow linearly) as the number of workers increases, especially when the workers are across different nodes.
I thought that MPI_scatter/MPI_gather is a parallel process, and wonder what leads to the above increasing? Is there any trick to make it faster, especially for workers distributing across CPU nodes?
The root rank has to push a fixed amount of data to the other ranks. As long as all ranks reside on the same compute node, the process is limited by the memory bandwidth available. Once more nodes become involved, the network bandwidth, usually much lower than the memory bandwidth, becomes the limiting factor.
Also the time to send a message is roughly divided in two parts - initial (network setup and MPI protocol handshake) latency and then the time it takes to physically transfer the actual data bits. As the amount of data is fixed, the total physical transfer time remains the same (as long as the transport type and therefore the bandwidth stays the same) but more setup/latency overhead is being added with each new rank that data is scattered to or gathered from, therefore the linear increase in the time it takes to complete the operation.
How an MPI_Scatter/Gather will work varies between implementations. Some MPI implementations may choose to use a series of MPI_Send as an underlying mechanism.
The parameters that may affect how MPI_Scatter works are:
1. Number of processes
2. Size of data
3. Interconnect
For example, an implementation may avoid using a broadcast for very small number of ranks sending/receiving very large data.
My understanding is that Bandwidth delay product refers to the maximum amount of data "in-transit" at any point in time, between two endpoints.
The thing that I don't get is, why multiply bandwidth by RTT. Bandwidth is a function of underlying medium, such as copper wire, fire optics etc and RTT is function of how busy intermediate nodes are, any scheduling applied at the intermediate nodes, distance etc. RTT can change, but bandwidth for practical purposes can be considered as fixed. So how does multiplying a constant value (capacity aka bandwidth) by fluctuating value (RTT) represents total amount of data in transit?
Based on this, will a really really slow have very large capacity? Chances are the "Causes" of RTT will start dropping.
Look at the units:
[bandwidth] = bytes / second
[round trip time] = seconds
[data volume] = bytes
[data volume] = [bandwidth] * [round trip time].
Unit-wise, it is correct. Semantically,
What is bandwidth * round trip time? It's the amount of data that left the sender before the first acknowledgement was received by the sender. That is, bandwidth * round trip time = the desired window size under perfect conditions.
If the round trip time is measured from the last packet and the sender's outbound bandwidth is perfectly stable and fully used, then the measured window size exactly calculates the number of packets (data and ACKs together) in transit. If you want only one direction, divide the quantity by two.
Since the round trip time is a measured quantity, it naturally fluctuates (and gets smoothed out). The measured bandwidth could fluctuate as well, and thus the estimated total volume of data in transit fluctuates as well.
Note that the amount of data in transit can vary with the data transfer rate. If the bottleneck is wire delay, then RTT can be considered constant, and the amount of data in transit will be proportional to the speed with which it's sent to the network.
Of course, if a round trip time suddenly rises dramatically, the estimated max. amount of data in transit rises as well, but that is correct. If there is no accompanying packet loss, the sliding window needs to expand. If there is packet loss, you need to reconsider the bandwidth estimate (and the bandwidth delay product drops accordingly).
To add to Jan Dvorak's answer, you can think of the 'big fat pipe' as a garden hose. We are interested in how much water is in the pipe. So, we take its 'bandwidth' i.e. how fast it can deliver water, which for a hose is determined by its cross-sectional area, and multiply by its length, which corresponds to the RTT, i.e. how 'long' a drop of water takes to get from one end to the other. The result is the volume of the hose, the volume of the pipe, the amount of data 'in the pipe'.
First, BDP is a calculated value used in performance tuning to determine the upper bounds of data which could be outstanding/unacknowledged. This, almost always, does not represent the quantity of "in-transit" data, but a target which tuning parameters are applied. If it represented "in-transit" data, always, there would be no room for performance tuning.
RTT does in fact fluctuate. This is why the expected worse case RTT is used in calculations. By tuning to the worse case, throughput efficiency will be at maximum when RTT is poorest. If RTT improves, we get outstanding Acks sooner, the pipe remains full and maximum throughput (efficiency) is maintained.
"Full pipe" is a misnomer. The goal is to keep the Tx side full, as the Rx contains Ack packets which are typically smaller than the transmitted packets.
RTT also aggregated asymmetrical upstream and downstream bandwidths (ADSL, satellite modem, cable modem, etc.).