Hot answers tagged

34

As others have pointed out, Google DNS uses IP anycast, which allows multiple servers in multiple locations to effectively share an IP address. Google (like many others) has many servers around the world that will respond to 8.8.8.8. So when you ping that address, the server that responds is the one closest to you, perhaps one in your country. Note that ...


24

Network latency is how long it takes for something sent from a source host to reach a destination host. There are many components to latency, and the latency can actually be different A to B and B to A. The round trip time is how long it takes for a request sent from a source to a destination, and for the response to get back to the original source. ...


23

My colleague believed it's because of the physical distance while I don't think it matters. My understanding is once you have done the initial handshake and the data flow has started, it doesn't matter where the server is located and the result should be almost the same. Am I missing something here? How does it really work? Both of you were right at some ...


19

Google in particular uses distributed datacenters around the globe. They announce the same IP network at various places and due to the way routing protocols work, you reach the nearest one. This is called anycast bbc.co.uk points to an IP address that belongs to Fastly, Inc, a content delivery network, that also has points of presence around the world, ...


16

You're overthinking. The number of lanes used doesn't really matter. Whether you transport 50 Gbit/s over 1, 2, or 5 lanes, the serialization delay is 20 ps/bit. So, you'd get 5 bits every 100 ps, regardless of the lanes used. Splitting of data into lanes and recombining takes place in the PCS sublayer and is invisible even on top of the physical layer. ...


14

The part that does the division to multiple lanes is called Physical Coding Sublayer in IEEE 802.3ba standard. This presentation by Gary Nicholl gives a good overview of it. The short explanation is that the data is divided to multiple lanes in blocks of 64 bits each (encoded on wire as 66 bits for clock recovery). Therefore as soon as packet size exceeds N*...


13

Neither, really. Replacing a copper link with a fiber link might lower latency a tiny bit (assuming an uncongested link), but what you really get when you replace a "core" link with a higher bandwidth link is less possible congestion. In your example scenerio, it doesn't matter, because there is only one device on each end. In a live network, though, moving ...


13

No, they won't for typical internet connectivity. You won't even be able to measure it using typical tools. For DC environments, we're fighting for every microsecond, but for typical ethernet switches, you get from 5-7 microseconds to 30-50 microseconds per hop (switch). Even 30-50 microseconds is too low to notice; typically latency is problematic for ...


12

Bell is telling you the truth. When you try to push 5Mbps (or more) into a 5Mbps connection, everything files into a neat little order (read: queue.) Your ping goes out without delay because there's no backlog. The reply, however, is now at the end of the queue. TCP is doing exactly what it's supposed to here -- the sender is filling the allowed receive ...


12

On Cisco devices, you can use Cisco IP SLA. You need to first configure and enable it, and then monitor the results. Steps: 1. ip sla monitor operation-number 2. type echo protocol ipIcmpEcho {destination-ip-address | destination-hostname} [source-ipaddr {ip-address | hostname} | source-interface interface-name] 3. frequency seconds 4. ip sla monitor ...


12

Hi and welcome to Network Engineering. As for "delay" vs "latency": The terms are not always used consistently. Some hints may be found here. I think generally, the term latency is used when looking at end-to-end times for one direction, which essentially are composed of the sum of all propagation, serialization, buffering (and possibly processing) delays ...


11

An an example Juniper MX80 has ingress->egress delay of about 8us, on low-latency cut-through switch it can be <1us (maybe 0.7us). (Remember that cut-through switch can't do cut-through 100% of time, only when egress port happens to be idle!) 1km in fibre is about 5us latency (again, single direction). Serialization delay @10G for minimum sized payload (...


11

Signal travels roughly same speed in copper and fibre, copper being slightly faster. Fibre is determined by refractive index, typically speed being about 0.65c, i.e 200km per 1ms (single direction, not RTT). Maximum theoretical throughput is hard to determine, maybedivide wavelengths to Planck length difference and light each up, so absurd amount. But in ...


10

This sounds like some form of "bufferbloat", probably on the part of the DSLAM/LNS that's performing the 6Mb rate limiting. It might be your CPE box, but that's a little less likely.


10

"Latency" can mean different things. Generally, it's a delay of some sort - application latency is the reaction time of an application (from input to output), network latency the delay for getting a packet from point A to B and so on. "Round-trip time" is more or less well defined as the network delay from point A to B and back . This is the sum of all ...


9

I would verify where the latency is occurring. Use a tool such as MTR which checks the latency at each hop. MTR combines ping statistics for each hop with a trace route, and can greatly help narrow down this type of problem. On a linux box the command would be mtr 8.8.8.8, there is also a windows version of this tool. The output will show you where the ...


9

One way to do this is ICMP Timestamp, which is milliseconds from midnight UTC. It has the added benefit that you don't necessarily need to control both ends, as long as the far-end is not firewalled, there is good chance it'll work. However, to have reliable one-way measurements, you need reliably same time in both ends. As ICMP timestamp only have ...


9

Does increasing the bandwidth on a link from lets say 1mb to 30mb reduce the RTT? In short, yes; you are changing serialization delay; at 1Mbps the serialization delay is non-trivial. Compare the serialization delay for a 1500 Byte packet at 1Mbps and 30Mbps: 1500 Bytes * 8 bits/Byte / 1,000,000 bits/second = 12 milliseconds (at 1Mbps) 1500 Bytes * 8 ...


9

Latency is the amount of time it takes a packet of data to leave your computer and receive a response back from the end point. That is why this is measure in time. This is key for time sensitive applications like VoIP and video conferencing. Bandwidth/data rate is the amount of data (bits) you can upload or download in a given time (seconds). This is key ...


8

If you're trying to test 1xGE No Drop Rate and measure circuit delay within 8ms, I would use nuttcp to test bandwidth and iperf2 / mtr to test delay. I would do the following... Find two linux desktop PCs, if possible (laptops are sometimes acceptable, but you might run into issues with chipset or bus performance at 1GE speeds). You can boot into a ...


8

Light travels down fiber at roughly 2/3 the speed of light. Therefore 420km should result in about 4-5 ms round trip. Also, keep in mind that 420km of aerial distance between two points does not necessarily mean that the fiber distance is 420km, it could be quite different. Depending on how many hand-offs/hops are on the line I would say that 15 ms round ...


8

Twisted pair cable used for the local loop has a velocity factor of about .58 - each km of cable takes ca. 6 µs to travel, adding ca. 12 µs to latency. The rest of the Internet probably uses fiber with a VF of .67, resulting in ca. 10 µs or .01 ms latency per km. ADSL has a basic encoding latency of around 10 ms. However, your ISP may likely interleave DSL ...


8

Yes, a bridge / switch adds some delay to a frame - in the order of 1 to 20 µs. For switches you usually speak of latency - the delay between receiving a frame and forwarding it out another port. A switch requires some time to receive the destination address and make the forwarding decision. Store-and-forward switches (the common kind) need to receive the ...


8

A cable half-way around the globe has a minimum latency of 100 ms, 200 ms round-trip (20,000 km distance / 200,000 km/s signal speed) - that's the physical limit.[1] In reality, links aren't as the crow flies (at all) and there are additional, active components in between, adding to latency - a more realistic figure is 150 or more one-way. Anything with ...


8

IP geolocation databases essentially represent a geolocation company's guess as to where an IP is located. Often they are based on the addresses in whois registrations, sometimes supplemented with the geolocation company's own research. For end-user IPs they are usually accurate to country level but for infrastructure IPs even that is frequently off. When ...


7

The latter half of your question seems to indicate you're looking for latency figures that take into account the process of forming application-layer data, in which case "ping" will not help much considering there's not much data to be formed in a ping packet. Network folks usually rely on ping because it's a relatively light and reliable way to produce a ...


7

One option would be to use IP SLA set up to send udp packets of the appropriate size / codec / etc. This can provide you back tons of information on the quality of your network, MOS score, etc. This is a reasonable article describing one way of setting up IP SLA feature that is relevant to voice diagnostics: http://www.networkworld.com/community/node/...


7

The question mentions that this is not hardware specific, however different platforms have different switching latencies and switching modes. Cut through switching is the fastest but lets through fragments. Fragment free is the second fastest and makes sure that frames are at least 64 bytes meaning they are not runts which is one of the signs of collisions. ...


6

If you have PCs at both ends then you could run xjperf, Qcheck from Ixia or other tools. You might get different results depending on if you use UDP or TCP and the number of sessions. For a distance over 100 miles you are looking at a minimum RTT of 1.6 ms at the speed of light in fibre/copper. So your RTT should be very low, maybe only a couple of ms. Say ...


6

iperf can do that job. Just ensure you test your devices back to back first so you know the capabilities of the devices. Of course there are more professional tools as well. This is a good RFC to read up proper methology: http://www.ietf.org/rfc/rfc2544.txt


Only top voted, non community-wiki answers of a minimum length are eligible