Most web Service suppliers advertise their performance in terms of downstream outturn. The “speed” that one pays for reflects, effectively, the quantity of bits per second which will be delivered on the access link into your home network. though this metric is sensible for several applications, it’s only 1 characteristic of network performance that ultimately affects a user’s expertise. In several cases, latency are often a minimum of as necessary as downstream outturn.
For example, think about the figure below, that shows online page load times as downstream outturn will increase—the time to load several websites decreases as outturn increases, however downstream outturn that’s quicker than concerning sixteen Mbps stops having any impact on online page load time.
The offender is latency: for brief, little transfers (as is that the case with several internet objects), the time to initiate a transmission control protocol association and open the initial congestion window is dominated by the round-trip time between the shopper and therefore the internet server. In alternative words, the dimensions of the access link now not matters as a result of transmission control protocol cannot increase its causation rate to “fill the pipe” before the association has completed.
The role of latency in internet performance isn’t any secret to anyone WHO has frolicked finding out it, and plenty of content suppliers together with Google, Facebook, et al. have spent sizeable effort to scale back latency (Google contains a project referred to as “Make the online Faster” that encompasses several of those efforts). Latency plays a task within the time it takes to finish a DNS operation, the time to initiate a association to the server, and therefore the time to extend TCP’s congestion window (indeed, students of networking can keep in mind that transmission control protocol outturn is reciprocally proportional to the round-trip time between the shopper and therefore the server). Thus, as outturn continues to extend, network latency plays Associate in Nursing more and more predominant role within the performance of applications like the online. Of course, latency additionally determines user expertise for several latency-sensitive applications in addition, together with streaming voice, audio, video, and gaming.
The question, then, becomes a way to cut back latency to the destinations that users unremarkably access. Content suppliers like Google et al. have taken many approaches: (1) putting internet caches nearer to users; (2) adjusting TCP’s congestion management mechanism to start out causation at a quicker rate for the primary few spherical visits. These steps, however, square measure solely a part of the story, as a result of the network performance between the online cache and therefore the user should still suffer, for a range of reasons:
First, factors like buffer bloat and telephone circuit interleaving will introduce important latency effects within the walk. Our study from SIGCOMM 2011 showed however each access link configuration and a user’s selection of kit (e.g., telephone circuit modem) will considerably have an effect on the latency that a user cannabis.
Second, a poor wireless network within the home will introduce important latency effects; generally we tend to see that two hundredth of the latency for real user connections from homes is at intervals the house itself.
Finally, if the online cache isn’t near users within the initial place (e.g., within the case of developing countries), the ways between the users and their destinations will still be subject to important latency. These factors are often notably evident in developing countries, wherever poor peering and interconnection may end up in long ways to content, and wherever the overwhelming majority of users access the network through mobile and cellular networks.
In the walk
In our SIGCOMM 2011 paper “Broadband web Performance: A read from the Gateway” (led by Srikanth Sundaresan and director Diamond State Donato), we tend to discerned many aspects of home networks which will contribute considerably to latency. we tend to outline a metric referred to as last-mile latency, that is that the latency to the primary hop within the ISP’s network. This metric captures the latency of the access link.
We found during this study that last-mile latencies square measure typically quite high, varied from concerning ten ms to just about forty ms (ranging from 40–80% of the end-to-end path latency). Variance is additionally high. One would possibly expect that variance would be lower for telephone circuit, since it’s not a shared medium like cable. amazingly, we tend to found that the other was true: Most users of cable ISPs have last-mile latencies of 0–10 ms. On the opposite hand, a big proportion of telephone circuit users have baseline last-mile latencies over twenty ms, with some users seeing last-mile latencies as high as fifty to sixty ms. supported discussions with network operators, we tend to believe telephone circuit corporations is also enabling Associate in Nursing interleaved native loop for these users. ISPs change interleaving for 3 main reasons: (1) the user is much from the DSLAM; (2) the user contains a poor quality link to the DSLAM; or (3) the user subscribes to “triple play” services. Associate in Nursing interleaved last-mile information path will increase strength to line noise at the value of upper latency. the value varies between 2 to fourfold the baseline latency. Thus, cable suppliers generally have lower last-mile latency and disturbance. Latencies for telephone circuit users could vary considerably supported physical factors like distance to the DSLAM or line quality.