What happens in an Internet minute?

Intel recently published a nice infographic slicing one minute of Internet traffic.

According to the infographic:

  • Aggregated Internet throughput is 85Tbps (639,800 GB per minute)
  • Pandora pumps 704Gbps (61,141 hours of music per minute, assuming 192Kbps)
  • Flickr pushes 530Gbps (20 million photos, assuming 200KB average photo)
  • YouTube generates 1.3Tbps (1.3 million video views, assuming 1Mbps average bit-rate)

I have some question marks regarding the figures. Traffic reports from various sources – Sandvine, DiViNetworks and others – show that in terms of throughput, YouTube comprises 5-15% of the overall traffic, whereas Pandora and Flickr do not make it to the major traffic contributors list at all.

Are all HTTP objects created equal?

You already know that we like to take close look at our traffic to improve our large and growing network. We also share our findings in the global data flow report. This time we analyzed HTTP objects flowing through our network.

DiViNetworksHTTPObjectsTrafficBySourceThe graph shows the percentage of overall HTTP objects flowing downstream in our network. Reminder – our network delivers international data capacity from 15 major Internet exchange points to over 100 ISPs in regions where capacity is expensive.

It’s interesting to compare the HTTP objects mix to the capacity volume mix, as we presented in September 2012.

The winner of the HTTP objects contest – Facebook, reaches 4-6th when it comes to volumes. CDNs such as Limelight and Edgecast, contributing significant volumes, apparently do that with much larger objects, hence do not appear on the HTTP objects map.
Dropbox, on the other hand, is an HTTP pump but at relatively low throughput volumes. My guess – HTTP graph provides early hints to the rise of new content pumps. As Dropbox free storage quota grows, we will see more and more traffic from their AS.
Want another hint? I was very surprised to hear today that Amazon’s traffic is balanced. Yes, their incoming traffic is as big as their outgoing. How come? Perhaps it has to do with Dropbox stored on Amazon’s S3.

How big is the developmental impact of broadband?

Broadband Internet access and economic prosperity go hand in hand. Since many of our ISP customers are in emerging markets we were wondering as to the correlation between the two.

We used SpeedTest‘s household download speed, Wiki‘s Internet penetration and the World Bank‘s GDP per capita.

Correlating the connected-household speed with the GDP (top graph below) you find that every 1Mbps correlates to $1,000 GDP per capita. When taking the Internet penetration into consideration, i.e. the average speed (bottom graph below), every 1Mbps correlates to $7,000 GDP per capita.

GDP_vs_ConnectedDownloadSpeedGDP_vs_AvgDownloadSpeedThe demand on the international links is linear to the average household download speed. The higher the penetration and the wider the access network are, the larger the load is on the international links feeding the country’s demand for capacity.

Assuming bandwidth and GDP are not only correlated, but also causal – economies can grow by finding affordable means to expand their international capacity.

We are happy to offer international capacity at half the market price and take part in accelerating economies growth.

Packet Loss – International and Access – Asia, LATAM, Africa

Handling over 100Gbps for over 50 ISPs worldwide, combined with smart devices on international and domestic PoPs, the DiViCloud Network holds a unique view on packet flow in international networks.

We recently are taking a closer look at packet loss on international and domestic links. Surprisingly, high packet loss occurs also during off-peak hours, without any congestion on the line – see results below. Such packet loss results in compromised QoE.

Packet loss – International and Access – DiViCloud’s Hong Kong PoP – Serving S.E.Asia

Packet loss – International and Access – DiViCloud’s NYC PoP – Serving LATAM

Packet loss – International and Access – DiViCloud’s Amsterdam PoP – Serving Africa

Summary of the above findings is presented below.

Packet Loss on International Links as Measured by DiViCloud

Holding both sides of the link, DiViNetworks can almost eliminate such packet losses, significantly improving end-users QoE.

An interesting case occurred this week, when a carrier providing transport from Singapore to one of our customers, suffered fiber fault. The following figures present the packet-loss rate prior to the fault, and during the fault.

Packet loss – International and Access – DiViCloud’s Amsterdam PoP – Serving Africa

Packet loss – International and Access – DiViCloud’s Singapore PoP – Serving S.E.Asia – During Fiber Fault

Additional observations are offered in our Global Data Flow Report.

Packet loss in International vs. Access networks in LATAM and Africa

As previously discussed, our DiViCloud global network of smart PoPs provides us with tons of insight about what’s going on in the big-Internet. We have shared some with you in our Global Data Flow Report.

This time we are taking a look at packet loss. High packet loss significantly compromises end-users’ experience and deteriorates link utilization. The reason being that packet loss signals to the TCP peers that the link is close to full, in which case, being a polite protocol, TCP reduces transmission rate.

The graphs below present packet loss in the Access network vs. packet loss on the International link. The measurements are taken on the DiViCloud PoP in London, serving Africa, and the DiViCloud PoP in NYC, serving LATAM.

Retransmissions in Access vs. International networks, as measured by DiViCloud

Clearly, such packet loss rates contribute to poor QoE. We’ll keep you posted about novel ways to eliminate this yellow-blue stuff, for the favor of the ISPs, content providers and CDNs alike.

Live traffic stats during El Clásico

This time we took a closer (and geeky) look at traffic stats during El Clásico - the major Spanish football match. We analyzed traffic flowing from our DiViCloud PoPs to one of the European ISPs we serve.

We isolated the Live traffic from the general traffic, by using the DiViLive capability. DiViLive identifies Live unicast sessions, transferring the same data to multiple users at the same time, and eliminates this redundancy, making way for x5-10 Live sessions.

The diagram below shows only the Live traffic during the period 19:00-01:00 on different days.

The El Clásico match took place on October 6th, demonstrating a spike of over 6x above the baseline live traffic volumes.

It was very interesting to see that in many other days, apart from October 6th, Live traffic spikes during evening hours. Can you match these to other Live events?

How did we get access to data flows of 100 ISPs worldwide?

Our Global Data Flow Report became very popular, with hundreds of downloads and shares. We received many questions about the report and its observations. The #1 question had to do with the way we gained access to the data.

Our DiViCloud network serves ISPs worldwide with virtual capacity. DiViCloud comprises of PoPs (points of presence) located at major Internet junctions.

Close to 100 ISPs are served by the DiViCloud network. 100% of the international downstream traffic to the served ISPs passes through these PoPs. As traffic passes through the PoPs, the software in the PoPs collects traffic statistics. These statistics serve as the raw data for the current report.

The information gathered includes: Originating AS (autonomous system), Destination AS, Traffic volume and Bit-stream repetition. We analyzed this raw information to obtain the analyses in the report, such as the one below.

Please share you observations or questions about the Global Data Flow Report.

CDN Contribution to Int’l Traffic in Different Regions

How much redundant traffic is left over by CDNs and Caching?

Fascinating to note that even in the presence of local CDNs and local Caching, international traffic still includes very high redundant traffic, passing over and over again on those expensive links.

International traffic repetition-rate by AS in ISPs with and without local caching

To gain access to the complete Global Data Flow Report – September 2012 please click here.

Related Infographic:

How repetitive is the Internet? (Infographic)

Our Global Data Flow Report Is Now Available

As our earlier blog posts here and here, we discussed global data flow and how we would publish our findings in DiViNetworks’ Global Data Flow Report. DiViNetworks holds a unique point of view – global PoPs positioned at strategic Internet junctions, combined with intelligent software overseeing masses of international traffic. This position enables us to study global data flow characteristics and trends. We would like to share these observations with you.

Below is an example of one of the graphs illustrated in the report to show originating AS source in different regions.

Originating AS (autonomous system) source in different regions

If you would like to gain access to the complete Global Data Flow Report – September 2012 please click here.

Slowdown in global Internet growth – trend or hiccup?

According to Telegeography’s Alan Mauldin, Global Internet capacity reaches 77Tbps, yet the growth is in decline due to a proliferation of CDNs and Cache.

The observation is indeed true, but it should be noted that the entry of a CDN or local Caching has merely a one time impact on growth ratios.

The analysis below simulates a single network operator, with some basic growth assumptions. Once the CDN and/or Caching solution is introduced into the network, a singular deceleration of international traffic is witnessed. In the following year demand reverts back to the organic pace – which is the product of broadband penetration and broadband speed.

The deceleration observed by Telegeography is a result of gradual introduction of CDN and/or Caching solutions . As such, systems will cover the market growth and return to its organic pace.

Want to learn more about Global Traffic? Follow our LinkedIn for or you can go here and fill out a form to be notified when the Global Data Flow report is available.