it here
Transcription
it here
Why is the Internet slow? Reasons and remedies for a “slow Internet” experience WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Why is the Internet slow? At some stage or other, everyone has heard that plaintive cry from a colleague, a family member, or indeed you’ve uttered it yourself. It’s also a complaint familiar to customer support desks at Internet Service Providers (ISPs) across the planet. Network ACCESS NETWORK SNRM, QLN, Crosstalk, Interleaving Depth Device Home Environment Hardware capabilities OS Browser Browser caching policy Security software DNS configuration Flash Background networkusing applications Figure 1: Content Modem/Router Ethernet, Powerplug or WiFi connection Latency to target node No. of objects per page WiFi configuration No. of domains per page DNS configuration Site dependencies Caching policy HTTP/2 support CDN Numbers and types of devices on WiFi and their current use Repetitive Electrical Noise Dry joints Intermittent RF noise Network Latency Network Contention Network policies NETWORK SERVICES DNS PARTNER NETWORKS Peering partners Network Congestion Network Routes TCP/IP scaling Some potential causes of a “slow Internet” experience Figure 1 highlights some potential issues that impact the user experience. We look at the technology, environmental and device choices that impact the user experience. We examine fundamental technology barriers that limit throughput and highlight why the user experience on a 100Mbps Internet connection isn’t 10 times better than a 10Mbps connection and the advent of 1Gbps connections won’t result in a 10-times better experience again. Finally, we provide a number of suggested actions for you to determine why the Internet appears slow to you and, where possible, to resolve same. Michael Slevin DIRECTOR, SONALAKE 1 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Let’s look at how it all works and what can cause a slow user experience Interactions between browsers (Firefox, Chrome, Safari, Internet Explorer and the like) and web sites generally use the HTTP protocol to retrieve or send multiple pieces of information, both text and images from a remote web site. These protocols in turn sit on top of the TCP/IP protocol. TCP provides reliability and importantly changes its throughput in reaction to packet loss, packets received out of order, or any similar reason to avoid having to re-transmit information. Each packet of information sent from A to B is acknowledged. The time taken for information to get from A to B and for the acknowledgement to get back to A is referred to as the round trip time (RTT) and is measured in milliseconds (ms). Today’s web pages can present data from multiple sources, be they advertisements, images and similar. Good practice for web developers calls for <40 round trips required per page. However the current average webpage is 2.2MB and only 27% of pages need <50 requests to correctly render the page1. On average 40 TCP/IP connections have to be established to load a page (although 30% of webpages need more than 40), with the average webpage including data from 17 domains (a max of 52 domains being observed)2. A result of this is that most of us at some stage will encounter web pages that hang … frequently caused by an advertisement host being down (planned or unplanned). Today’s browsers such as Firefox, Chrome support 6 simultaneous HTTP connections to a single domain which introduces a degree of parallelism, but doesn’t help with the multiple domains per web page issue3. Customer Experience Distance, Obstacles & Orientation? 100Mbps or 1Gbps ethernet port? Machine & drivers capable of powering 1Gbps Operating System, Machine efficiency WiFi is contended; Neighbouring WiFi interferes Browser version? IE, Chrome, Firefox, Safari Browser cache WPA2 better/faster than WPA; One or multiple Encryption modes supported? Flash version; Security/antiVirus software DNS config; Orientation? Environment interferes; Distance, walls, insulation a factor DNS configuration RF Noise? Internal house wiring correct? Figure 2: Summary of in-home/on-device issues impacting throughput 1 http://httparchive.org/interesting.php (March 2016) 2 http://httparchive.org/trends.php (March 2016) 3 HTTP/22 was introduced in 2015 and supports multi-host multiplexing. As this is adopted more, users should experience an improvement in page load times. ~7% of websites supported HTTP/2 over TLS in February 2016. 2 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Don’t forget too, that rendering a single web page will also require a potentially large number of DNS lookups (to convert the URL to an IP address). So the choice of DNS server is important. The default servers are usually those provided by your ISP and the default latency is similar to the RTT for the access network. There are other DNS providers such as Google and OpenDNS which can be used, but don’t select servers that have large RTT. DNS servers do give trouble from time to time with the impact seen as timeouts or “slow” web page rendering for a large proportion of the ISP’s customers. It is usually possible to select a primary and secondary DNS – use your ISPs as primary and another as secondary. Now let’s look at the networks ADSL2+ access networks seem to have typical RTTs of ~20-30ms. VDSL and Cable networks seem to be in the 6-10ms range while FTTH networks are in the 3-6ms range. 3G mobile networks can have latencies >100ms while 4G is >50ms4. For mobile-facing websites, the number of round trips to render a page should be less than 5 to account for the longer latency. All of these relate to the latency from the modem to the nearest commercial sites. If the two end points are 1ms from each other, we can achieve a max throughput for a single TCP/IP session of 509Mbps with the default TCP/IP window size of 65,5355. At 10ms, it’s 51Mbps. Yes that’s right, a single TCP/IP session can’t deliver more than 51Mbps when the two end points have a 10ms RTT between them. By the time you get to 30ms, you’re at 17Mbps maximum download for a single TCP/IP session. We can see from the above that it’s actually RTT that has a greater impact on performance than the available bandwidth. An individual on a typical VDSL or cable 8ms network offering 100 or 300Mbps on a default Windows machine can never see more than ~64Mbps for a single connection to a very local resource. Incidentally, your uplink path is as important as your downlink path, since in a congested scenario, your acknowledgement (ACK) of a received packet is what triggers the transmission of the next. If that gets lost, or a timeout occurs, we see data retransmissions, duplicate ACKs, and a consequent slowdown in the achieved throughput. 4 Ofcom found an average latency of 53ms for 4G and 63ms for 3G networks in Q4 2014 (http://stakeholders.ofcom.org. uk/market-data-research/other/telecoms-research/broadband-speeds/mobile-bb-april-15/) 5 Note that for all these examples, the TCP window size is assumed to be 65,535 bytes. This can be increased through TCP/IP window scaling. This is enabled by default on Mac computers, but, while supported in Windows, its use is usually restricted when transiting public networks. To enable scaling on a Windows 7 machine, try the command “netsh interface tcp set heuristics disabled” from a command line with administrative privileges. Note that it will only make a difference where latency is long and your bandwidth is high. Check your current TCP/IP settings at www.speedguide.net/ analyzer.php TCP throughput (in bps) = TCP/IP window in (bits) / RTT (seconds) (but we can also consider that only ~97% of this is payload, the remainder being TCP/IP header) 3 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE All Internet access technologies are sold on a contended basis. Unfortunately, the assumptions underlying the contention ratios may no longer be valid. 300 people attached to a single DSLAM with an aggregate 2.4Gbps of bandwidth available to them will encounter problems at peak times getting off that DSLAM which may only have a single 1Gbps interface into the network. Traffic volumes are growing, leading to pressures on capacity planning and deployment in the access and backhaul network. Figure 3 illustrates temporary network congestion6 experienced at two sites connected to the same DSLAM over the Christmas/New Year break. The graph shows minimum TCP/IP RTT experienced between test devices and a test location. The normal RTT is 15ms, but during times of congestion, the minimum (not average) RTT observed during test transactions on one day exceeded 200ms. The impact to a customer is an experienced throughput as low as 1.5Mbps versus the normal 13Mbps (limited by the ADSL2+ bandwidth). Figure 3: Indications of network congestion around the Christmas/New Year period (VisiMetrix dashboard of Visualware generated data) OK, so those metrics relate to the network between modem and the ISP. However, in the case of fixed line Internet access, most devices will connect via WiFi. Most readers will know that WiFi uses unlicensed radio spectrum and thus is subject to interference leading to packet loss. WiFi comes in many flavours with maximum rated line speeds of 11, 54, 54-600 and 866Mbps (for 801.11b, g, n, ac respectively). However, that 54Mbps for 802.11g refers to the physical bit rate which translates to a ~22Mbps data rate when error correction overheads are taken into account. Some WiFi routers permit configuring WiFi in a mode compatible with multiple variants, e.g. 802.11b/g. Having just one device operating at the lower/older standard impacts everyone. Of course, this bandwidth is shared among connected devices. WiFi operates based on principles of collision avoidance. This works well on lightly loaded networks, but deteriorates quickly once you get above 10% loading. 6 We can infer congestion with confidence in this instance because we observed the same pattern on a small number of test lines while the remainder were not impacted. 4 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Radio interference can originate from neighbouring WiFi networks, game controllers, wireless video cameras, older Bluetooth devices plus the usual set of RF noise sources including fluorescent lights. The reliability of the connection between your device and the WiFi router depends on multiple factors including the aforementioned noise, distance between the devices, radio reflections and movement leading to variable levels of packet loss as the radio environment changes – a simple change in the orientation of the antenna on the wireless device can have significant impact. Interference causes WiFi packet loss which causes WiFi retransmissions which mean greater variability in the arrival rates of the IP packets carried on the WiFi signal. This variability contributes to increased RTT (2ms to >60ms) and thus reduced TCP/IP throughput rates. WiFi repeaters/extenders are also bad news except on lightly loaded networks. The practical response to this is to ensure you have a new router and give yourself some breathing space by using channels in the 5Ghz band rather than the crowded 2.4GHz band. The configuration of the WiFi itself causes issues. There is huge variation in the performance of wireless routers, with some popular ones having appalling reputations. Now look at the device itself, be it laptop, smartphone or similar. The obvious things that impact Internet speed include old versions of browsers or very new versions on old hardware(!), the local caching policy on the browser, and anti-virus software. Buggy network drivers can impact. And of course, many users forget that they can have many potentially bandwidth hungry applications running in the background, examples being Dropbox or torrent client/servers. The devices themselves can vary hugely in the performance offered and more expensive does not necessarily mean better. For Windows users, clean up your machine with CCleaner, get rid of any nasties with Malwarebytes. Delivery of popular web content relies on Content Delivery Networks (CDNs). Essentially, these are local copies of content placed on servers within or close to a particular ISP (to achieve low latency). Use of an international DNS server can result in content being delivered from the source or from a CDN in another country rather than the local copy which is available much faster (larger bandwidth to the server and reduced RTT). 5 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Speed Tests But what about Internet speed tests? It can be very frustrating to have a poor Internet experience while a speed test will happily show that your Internet connection is 10, 20, 50, 100Mbps. Tests run against Ookla’s www.speedtest.net test servers will happily show you speeds greater than this. Ookla’s Speed tests are multi-threaded. They discard the fastest 10% (because they can show momentary b/w greater than your connection due to buffering within the network and device). They also discard the 30% slowest results. So the results shown are not indicative of what you as a user might expect at all – it really only indicates what your line is capable of in total. They are not measuring the quality of the connection (in terms of packet loss, or capability to sustain a high throughput). However, knowing the capacity of a link is important, particularly for people on the first generation ADSL technology. All of the copper-based fixed line access technologies rely on RF signals carried on the wire (coax or twisted pair) which makes them subject to electrical/ RF noise events. While ADSL2+ and VDSL lines will generally recover after an intermittent noise event, the original ADSL technology will not and requires you to restart your modem to regain your normal connection bandwidth. Very occasionally, an ADSL2+ line will revert to ADSL in reaction to a noise event, requiring a modem restart to get it back to ADSL2+. We have also seen VDSL lines train themselves down from the 40Mbps range to 3Mbps due to line impairments. So keeping an eye on your modem sync rate is worthwhile, either directly or by proxy using a capacity speed test. A single-threaded TCP/IP test enables us to look at the real user experience in a lot more detail. Figure 4 on the following page is a single threaded test showing TCP/IP throughput over an 8-second timespan on an ADSL line from a test server. You can see that a throughput of ~7.5Mbps to 8Mbps is achieved for most of the time. However, you can also see that there were three large pauses (1400ms, 500ms, 500ms) when no data was transferred. You can also see a single burst to near 10Mbps, all of this on a line synchronised at 7.4Mbps. www.speedtest.net would show this as a perfect line, but the detailed graph of a single session below shows significant problems. The packet loss highlighted in red, results in retransmissions of data, so the useful data transmitted in the 8-second window is much less than expected which impacts the performance of the application or user’s browsing experience. 6 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Figure 4: A single TCP/IP download over 8 seconds (Visualware public test) By contrast Figure 5 shows the millisecond-by-millisecond behaviour of a perfect session (VDSL line delivering 40Mbps over 7ms RTT). Figure 5: A near perfect single TCP/IP download over 8 seconds (Visualware non-public test) 7 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE The graph in Figure 6 illustrates a bad congestion scenario – lots of gaps where no data is transmitted resulting in an average throughput of 1.2Mbps instead of a normal 7Mbps. Figure 6: TCP/IP throughput over an 8-second period (congested) (Visualware non-public test) Incidentally, in a scenario like this, you will find that the minimum RTT for a set of packets will be impacted, so the congestion will manifest itself in a the minimum RTT for a PING response moving from say 12ms to 23ms in the scenario illustrated above. While not 100% reliable, PING response times can act as an indicator of congestion and packet loss in the network. As we’ve said above, Ookla’s www.speedtest.net will return the overall capacity of a line. It doesn’t tell you what you can usefully get out of that line in terms of sustainable downloads or in terms of reliable connectivity. www.speedof.me has a nice intuitive animated results screen shown below (but can cause Firefox and Chrome to crash). Figure 7: A sample www.speedof.me test result 8 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Testing from Ireland against a UK-based server gave us ~35Mbps which is respectable given the 25ms RTT to the server. Ookla’s www.speedtest.net returned ~51 Mbps to a server with 7ms RTT 7. www.testmy.net is interesting (46Mbps from UK-based server but no RTT given). It supports both single and multithreaded tests. The www.testmy.net test also responds correctly if you override windows settings and permit TCP/IP windows scaling on public networks. With TCP windows scaling enabled, a 5.9Mbps download throughput from Hong Kong to Ireland becomes 16.4Mbps. So use www.speedtest.net to see what the Internet connection is capable of and www.speedof.me & www.testmy.net to see what the connection can achieve with a degree of reliability. Visualware (www.myspeed.visualware.com) gives the most detail including jitter/packet loss for VoIP simulation, capacity tests and single threaded TCP/IP Quality tests. This latter test can be quite revealing but its efficacy is limited by the low number of test servers available. Visualware returned ~48Mbps capacity to a server with 13ms RTT. Single threaded TCP/IP quality tests to the UK reveal a rate of 14Mbps from a server with 28ms RTT (18Mbps is the theoretical max) and highlighting a lot of packet reordering and packet loss. We can get a sustained 3.7Mbps to a server in the USA with RTT of 121ms (which is close to the theoretical best we can get of 4.2Mbps2). As you can see, latency matters a lot – shaving 5 milliseconds from the London-New York route was the sole driver for the Hibernia Express sub-sea cable completed in 2015. Google’s Network Diagnostic Test (www.measurementlab.net/tools/ndt/) gives similar results and detail to Visualware. Use Visualware or NDT for deeper diagnosis if the results of the first tests above are poor – the tests are also useful to highlight issues with specific devices. 7 Speedtest tells me I have ~240Mbps download capability on my home network, even when run from a 10 year old laptop. This is primarily because the test is run against the ISPs own test servers only 7ms away. Speedof.me tells me I can get 14Mbps to London, but importantly tells me that I’m 47ms away from London on my home network. The 240Mbps capability is almost irrelevant in the face of such latency, and it is that higher latency that normal Internet usage is subjected to. In this instance, a lower bandwidth access technology with a headline 50Mbps download capability will in fact achieve much better results than the 240Mbps line due to lower latency. Your ISPs connectivity to the domains you’re interested in matters. 9 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Netalyzr (www.netalyzr.icsi.berkeley.edu) takes a while to run, but is superb, and available as a smartphone app for normal people or as a command line tool for readers of this paper8. It includes bandwidth tests, but a lot more besides, checking DNS settings, firewall. Note that the 18Mbps recorded in the test below (Figure 8) from a server 96ms away implies a degree of multithreading is occurring – 4 parallel sessions could drive 21Mbps throughput at 96ms. No packet loss occurred, so we might hypothesise that Netalyzer uses 4 parallel sessions. Figure 8: An extract from the Netalyzer test results We’ve seen that there are potentially many answers to the question “why is the Internet slow for me”? On the following page is a partial checklist to help diagnose your issues. Of course, determining the root cause of a perceived slow connection is always much easier if it impacts everyone at a particular site. It also helps if you’ve prepared a little bit and have collected some baseline measurements. From your modem, collect and store the current xDSL metrics, downstream and upstream sync rates for xDSL, SNR metrics and interleaving depth – just take a snapshot and store it – see Figure 9, Figure 10 and Figure 11 (for DOCSIS cable modems). Metrics that reflect the stability of your line include the number of spontaneous resynchronisations since the modem was powered up, the time since the last resynchronisation, the Errored Second (ES) and Severe Errored Second (SES) count. Not all modems report this information. 8 A sample test result is found at http://netalyzr.icsi.berkeley.edu/restore/id=example-session 10 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Figure 9: Principal ADSL metrics (from modem) Figure 10:DOCSIS metrics (from modem) 11 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Figure 11:DOCSIS SNR (from modem) Ping a couple of reference sites, local and international and store the results – better still use traceroute. Traceroute will tell you the RTT to the target site and also the route (and number of hops) taken to get there9. It will help to have these for comparison with the results you get the next time you feel that “the Internet is slow”. If the results are similar, don’t blame the ISP. The “MTR” application combines both ping and traceroute and may be a handier solution for you. We’ve seen that slow throughput is potentially due to a number of causes. 9 We have observed on a few occasions that throughput was being impacted every time we routed via a particular LNS (the first hop in the ISP’s network). You’ll only find that out by doing a traceroute at regular intervals so that you have a set of good and bad results. 12 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE The table below is a guide to help you determine whether the cause is the remote site you’re interacting with, the latency between your device and the remote website, your ISP’s peering links and network (including DNS servers), your access to the ISP, your home network or your device(s). While some remedies are offered, we haven’t the bandwidth here to cover all issues for all device types, nor for all modem and network types. Symptom What to do A few sites are slow – eBay, Amazon, some popular content you’ve seen on Facebook, Twitter It might simply be that a site is particularly busy, can’t cope with the traffic or is having problems. We seem to expect/ accept this for Ticketmaster but for very few other sites. Sadly, we’re also seeing more denial-of-service attacks which drastically impact targeted websites. It’s only one device on the network affected? – nobody else is complaining Check with other devices. Run a speed test from more than one device. Use Visualware, Netalyzer or NDT. Is everyone on site similarly impacted? For ADSL and ADSL2 users, the WAN bandwidth is likely to be the limiting factor – for VDSL, Cable and FTTH networks, either WiFi, or the two ends of the communication are likely to be the limiting factors. It might also be that some of your ISP’s (or another provider’s) peering links are saturated towards those sites. This is where it helps if you have pre-recorded some traceroute responses to compare the good times with the bad. If its just one device, focus on getting it back to “normal”. Possible issues could be anti-virus software, bad drivers, use of a different (bad) DNS server, stale or corrupt caches, browser add-ons or plugins, OS thrashing, duplicate LAN IP address allocation, etc. For Windows folk, use CCleaner as a first port of call. Run a speed test – www.speedtest.net to see what the “pipe” is capable of and www.speedof.me, www.testmy.net to see what a line can achieve with a degree of reliability. Check via your modem that you are still synced at your expected data rate e.g. 34Mbps. For xDSL, it’s the downstream and upstream line rate (typically given in kbps) so 34000 represents 34Mbps. Depending on the diagnostics, you might see how many times a modem has resynchronised (re-connected) since it was booted. This figure should be low, less than once a week. A very high figure (once or more per day) indicates some impairment on your line – a topic for another day. If your downstream/upstream attenutation has changed, there’s been a change in the physical copper line. Were you playing with the phones at the weekend? If not, then get your ISP to investigate. Is it an unusually dry or wet spell? 13 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Is everyone on site similarly impacted? (continued) High numbers of Errored Seconds (ES) or Severe Errored Seconds (SES) point to problems with the DSL line or your environment in terms of noise. A high interleaving depth introduces latency but with the benefit of a more stable connection. If your SNR is well above 6dB, consider asking your ISP to reduce the interleaving Depth. Kitz (www.kitz. co.uk) is a good source of information for those of you interested in learning more about DSL lines. If your SNR is poorer than before, your RF conditions on the line have changed – it might be intermittent or permanent (e.g. a neighbour has subscribed and your line is suffering from increased crosstalk). On some modems, you can see the Quiet Line Noise (QLN). Power cycle the modem and check that you return to your previous “normal” link speeds. Make sure your modem is correctly wired and DSL microfilters are in use where needed. Another problem which likely impacts all attached devices is the default DNS server for your home network. Try manually configuring a DNS server to Google’s (8.8.8.8) and check again. If your ISP’s DNS server is in trouble, try reconfiguring the DNS server on the modem. Compare your current DNS response times to others10. Another reason you might want to change DNS provider (other than speed) is related to restricting access to particular sites. For example Norton has specific free DNS servers which restrict access to adult themed sites and OpenDNS allows you to restrict specific sites. Still got problems? Connect a device to the modem by ethernet cable and run speed tests. Connect the same device via WiFi and retest compare your ethernet and WiFi experience. Ethernet results are bad? What else is running on your home network? Disable WiFi temporarily and remove other ethernet-attached devices. If you get a good result, bring back Wifi and try again on ethernet. A bad result would indicate that one or more devices on your network is hogging your bandwidth. Check - use the modem or download the “Fing app”. Don’t forget that your uplink is as important as the downlink. A congested or busy uplink will severely impact the throughput achievable on the downlink. Maybe its time to prioritise your traffic or a specific traffic type or traffic from a set of devices. Consult your router manual and configure & enable QoS. 10 GRC’s DNS Benchmark (www.grc.com/dns/benchmark.htm) or Namebench (https://code.google.com/archive/p/ namebench/downloads) will point you in the right direction. NOTE of course, that value added services such as VoIP and TV will likely preclude you from changing the DNS server settings on the modem. You can of course still change the DNS configuration for specific devices. 14 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE If you suspect that its WiFi performance Download a WiFi monitoring app and check signal strength at various locations. We like “WiFi Analyser” and “Netgear WiFi Analytics”. The latter can record strength in different rooms and also look at channel interference. A neighbour’s WiFi may be killing your WiFi in part of your house – the app will show you less utilised channels. If you want to be more thorough, download a WiFi mapping app which, with a bit of patience, can generate a WiFi signal strength heat map overlayed on a floor plan of your house. Caveat – no app will show you RF interference from sources other than other WiFi signals. Strengths look good? Check what devices are on WiFi – use the modem or download the “Fing app”. Again look at the WiFi configuration – see if you can find a simpler configuration. Teenagers tend to share WiFi passwords – it’s quite possible that the gang of teenagers hanging about outside your house are on your WiFi. Momentarily change the SSID of your WiFi, link to that and check speeds again. Good speeds now will point to WiFi congestion within the home. At this stage, we’re beginning to suspect congestion in the network, but is it the access network or selected peering links? All Internet access technologies are sold on a contended basis. Unfortunately, the assumptions underlying the contention ratios may no longer be valid. 300 people attached to a single DSLAM with an aggregate 3Gbps of bandwidth available to them will encounter problems at peak times getting off that DSLAM which may only have a single 1Gbps interface into the network. Traffic volumes are growing leading to pressures on capacity planning and deployment in the backhaul network. To “prove” congestion outside your home, take reference readings (minimum and average RTT times to and from a number of sites). Do this from both WiFi and Ethernet connected devices. MTR (or WinMTR) is a handy tool for this. Now run the same set of tests when you suspect a problem. The Minimum RTT should be never change by more one or two milliseconds (a route may have changed). The average can fluctuate depending on the busyness of the network. A change in the Minimum RTT or of course significant packet loss (for both WiFi and Ethernet connected devices) is a sign of congestion outside your control. 15 WHY IS THE INTERNET SLOW? REASONS AND REMEDIES FOR A “SLOW INTERNET” EXPERIENCE Sidebar Of course, having a regular history of line conditions enables you to identify changes in line condition and their impact, when they happen. Figure 12 shows the impact of an increase in Quiet Line Noise on the carrying capacity of an ADSL line - the increase in Quiet Line Noise (3rd graph) reduces the carrying capacity of the line (2nd graph). The carrying capacity is represented by the area under the lines in the 2nd graph – the area under the red line is ~15% of the area under the blue line, meaning that throughput is reduced to 15% of normal when the noise on the line increased. The root cause was a misbehaving modem power supply. Figure 12:Detailed snapshots of line conditions showing the “good” state in blue and the “bad” state in red. 16 SONALAKE, Quantum House, Temple Road Blackrock, Co. Dublin, A94 XOH9, Ireland [email protected] | +353 1 278 8750 www.sonalake.com Sonalake helps companies to innovate using software, from the biggest global Internet brands to the latest start-up disruptors. We provide product development, software engineering and expert consulting to help our clients capitalise on their know-how and data. Our track record of solution delivery is second to none, ranging from carrier-grade platforms for communication networks to advanced user interfaces for online business applications. Our technology focus is on cutting edge open-source software across the full stack, complemented by our agile approach and our platform frameworks that offer accelerated time to market. We apply our software expertise within multiple vertical industries, cross-pollinating new technology ideas and best practices. And we invest in research and innovation to stay ahead, building on our domain expertise and process capabilities. It all adds up to deliver the special customer experience that we have become known for. Sonalake was formed when DANU Technologies and Verax Systems merged in February 2016 following 10 years of close partnership. Our IT Management Solutions division remains branded as veraxsystems.com, supporting many clients and resellers globally.