ISP Information:
Applications or systems that are able to scale to large amounts of users. For example, a database that completely locks out every other user when someone is using it is NOT scalable. The computer system that runs ATM and bank transactions must be highly scalable. It is often misspelled as "Scaleable," and is used in many product names.
ISP Glossary:
Scalable - qazmlp wrote: Two nodes are connected in the same IP network. The average bandwidth of the IP link between those 2 nodes is 'T' MB/sec. These 2 nodes are 'D'(maybe, 200 or 300 km)km apart from each other. In that case, how much time it will take for transferring 'A' MB amount of data from one node to the other one? I am just confused about how the distance need to be considered for calculating this.It doesn't, really, because you specifically said an IP network andthe time it takes for a signal to travel three hundred kilometers(about one millisecond each way, IIRC) is only a small part of theround trip time that affects throughput.That throughput depends on the protocol you choose for two reasons:overhead and capacity utilization. Generally speaking, the overheadis small compared to the question of whether you're actually takingadvantage of the capacity of the link.For instance, if your transfer protocol is XMODEM, which requires youto send a 128-byte block of data and wait for an acknowledgementbefore continuing, and your round-trip time is 50 milliseconds, thenno matter how fast your link is or how little time it takes for eachend to send or receive data, compute checksums, and move on, you'resimply not going to get speeds in excess of 2,560 bytes/second (50milliseconds per exchange means no more than 20 128-byte blockstransferred per second.) At 300 or even 1200 bps, the time requiredto send 128 bytes of data made the pause during which nothing was sent(and capacity was wasted) very small and Ward Christensen was a hero.However, at even dialup speeds today, using XMODEM is a waste ofcriminal proportions and thus it is never used over high speed links;I describe it here only to illustrate the issue in ridiculous extreme.For things like web surfing, HTTP is used, and it in turn uses TCP.TCP has what is known as a "receive window", which defines how far asender should let transmission get ahead of the receiptacknowledgements it gets. If the acknowledgement of the first segmentof data gets back to the sender before the amount of unacknowledgeddata reaches the size of the receive window, then the sender neverstops transmitting and the throughput of the connection approaches thebandwidth of the link.To accomplish that, the receive window must be as large as (orslightly larger than) the amount of data that can be transmittedbefore the acknowledgement of the first segment gets back to thesender, which will be at least the round-trip time multiplied by theoverall link speed (possibly the speed of the slowest link.) Since itstill takes some time to send data, to read it, compare checksums,etc. the actual acknowledge time will be at least slightly greaterthan the ping time.Now for a possible bad side effect for the average user: the abovecalculation is for using the entire capacity for a single TCPconnection, such as an ftp transfer, and usually results in a prettylarge window. If you have a high-speed link such as cable or DSL andconfigure your TCP receive window to be that large, what do you thinkis going to happen when you have multiple connections open, as whenyou are downloading a couple of different files plus surfing the webplus getting your USENET fix? You've just told umpteen servers eachto blast you enough data to fill the slowest link (for most consumers,that's your last mile), and (ideally) the data will queue up at thebottleneck and increase the round-trip time for your link; no matter,you have multiple connections going so the fact that each might pauseintermittently is not going to affect your total throughput. However,no queue is infinite in capacity (even less so for consumer broadbandaccess servers) and eventually data will be dropped.Your receiver will notice sequences missing and requestretransmission, which will set your download back a bit ... and causeeven more data to be sent, and again some will be lost, etc. etc. etc.Generally users should think twice about tuning their TCP receivewindow to guarantee maximum throughput even from servers hu
|
