ICFA/SCIC Network Monitoring Report, February 2005

International Committee for Future Accelerators (ICFA)

Standing Committee on Inter-Regional Connectivity (SCIC)

Chairperson: Professor Harvey Newman, Caltech

ICFA SCIC Network Monitoring Report

Prepared by the ICFA SCIC Monitoring Working Group

On behalf of the Working Group:
Les Cottrell cottrell@slac.stanford.edu

January 2005 Report of the ICFA-SCIC Monitoring Working Group

Edited by R. Les Cottrell on behalf of the ICFA-SCIC Monitoring WG

Created January 10, 2005. Last Update February 7, 2005

ICFA-SCIC Home Page | Monitoring WG Home Page

This report is available from http://www.slac.stanford.edu/xorg/icfa/icfa-net-paper-jan05/

Executive Overview

Internet performance is improving each year with packet losses typically improving by 40-50% per year and Round Trip Times (RTTs) by 10-20% and, for some regions such as S. E. Europe, even more. Geosynchronous satellite connections are still important to countries with poor telecommunications infrastructure. In general for HEP countries satellite links are being replaced with land-line links with improved performance (in particular for RTT).

Links between the more developed regions including Anglo America, Japan and Europe are much better than elsewhere (5 - 10 times more throughput achievable). Regions such as S.E. Europe, Central Asia and Russia are catching up with the more developed regions (at the present rate of progress they should catch up by the end of the decade). However, China, the Middle East, Latin America and Africa are several years behind in performance compared to the more developed regions, and do not appear to be catching up. Even worse India appears to be falling further behind.

For modern HENP collaborations and Grids there is an increasing need for high-performance monitoring to set expectations, provide planning and trouble-shooting information, and to provide steering for applications

To quantify and help bridge the Digital Divide, enable world-wide collaborations, and reach-out to scientists world-wide, it is imperative to continue and extend the monitoring coverage to all countries with HENP programs and significant scientific enterprises. This in turn will require help from ICFA to identify sites to monitor and contacts for those sites, plus identifying sources of on-going funding support to continue and extend the monitoring.

Introduction

The formation of this working group was requested at the ICFA/SCIC meeting at CERN in March 2002 [icfa-mar02]. The mission is to: Provide a quantitative/technical view of inter-regional network performance to enable understanding the current situation and making recommendations for improved inter-regional connectivity.

The lead person for the monitoring working group was identified as Les Cottrell. The lead person was requested to gather a team of people to assist in preparing the report and to prepare the current ICFA report for the end of 2002. The team membership consists of:

Table 1: Members of the ICFA/SCIC Network Monitoring team

Les Cottrell	SLAC	US	cottrell@slac.stanford.edu
Richard Hughes-Jones	University of Manchester	UK	rich@a3.ph.man.ac.uk
Sergei Berezhnev	RUHEP, Moscow State.Univ.	Russia	sfb@radio-msu.net
Sergio F. Novaes	FNAL	S. America	novaes@fnal.gov
Fukuko Yuasa	KEK	Japan and E. Asia	Fukuko.Yuasa@kek.jp
Sylvain Ravot	Caltech	CMS	Sylvain.Ravot@cern.ch
Daniel Davids	CERN	CERN, Europe, LHC	Daniel.Davids@cern.ch
Shawn McKee	Michigan	I2 HENP Net Mon WG	smckee@umich.edu

Goals of the Working Group

Obtain as uniform picture as possible of the present performance of the connectivity used by the ICFA community
Prepare reports on the performance of HEP connectivity, including, where possible, the identification of any key bottlenecks or problem areas.

This report may be regarded as a follow on to the May 1998 Report of the ICFA-NTF Monitoring Working Group [icfa-98], the January 2003 Report of the ICFA-SCIC Monitoring Working Group [icfa-03] and the January 2004 Report of the ICFA-SCIC Monitoring Working Group [icfa-04]. The current report updates the January 2003 report, but is complete in its own right in that it includes the tutorial information from the January 2003 report. As such it is not very different from the January 2003 report.

Methodology

There are two complementary types of Internet monitoring reported on in this report.

In the first we use PingER [pinger] which uses the ubiquitous "ping" utility available standard on most modern hosts. Details of the PingER methodology can be found in the May 1998 Report of the ICFA-NTF Monitoring Working Group [icfa-98] and [ejds-pinger]. PingER provides low intrusiveness (~ 100bits/s per host pair monitored¹) Round Trip Time (RTT), loss, reachability (if a host does not respond to a set of 21 pings it is presumed to be non-reachable). The low intrusiveness enables the method to be very effective for measuring regions and hosts with poor connectivity. Since the ping server is pre-installed on all remote hosts of interest, minimal support is needed for the remote host (no software to install, no account needed etc.)
The second method (IEPM-BW [iepm]) is for measuring high network and application throughput between hosts with excellent connections. Examples of such hosts are to be found at HENP accelerator sites and tier 1 and 2 sites, major Grid sites, and major academic and research sites in Anglo America², Japan and Europe. The method can be quite intrusive (for each remote host being monitored from a monitoring host, it can utilize hundreds of Mbits/s for ten seconds to a minute each hour). It also requires more support from the remote host. In particular either various services must be installed and run by the local administrator or an account is required, software (servers) must be installed, disk space, compute cycles etc. are consumed, and there are security issues. The method provides expectations of throughput achievable at the network and application levels, as well as information on how to achieve it, and trouble-shooting information.

PingER Results

The PingER data and results extend back to the start of 1995. They thus provide a valuable history of Internet performance. There (January 2005) are about 38 monitoring hosts in 13 countries, about 55 Beacon remote hosts that all monitoring sites, and over 980 remote hosts that are monitored at 673 sites in 114 countries (see PingER Deployment [pinger-deploy]). These countries contain over 78% of the world's population and over 99% of the online users of the Internet. Most of the hosts monitored are at educational or research sites. We try and get at least 2 hosts per country to help identify and avoid anomalies at a single host. The requirements for the remote host can be found in [host-req]. Fig. 1 below shows the locations of the monitoring and remote (monitored sites).

Figure 1: Locations of PingER monitoring and remote sites as of Jan 2005.

Since there are over 3700 monitoring/monitored-remote-host pairs, it is important to provide aggregation of data by hosts from a variety of "affinity groups". PingER provides aggregation by affinity groups such as HENP experiment collaborator sites, Top Level Domain (TLD), Internet Service Provider (ISP), or by world region etc. The world regions are shown below in Fig. 2. They are chosen starting from the U.N. definitions [un]. We modify the region definitions to take into account which countries have HENP interests and to try and ensure the countries in a region have similar performance. The regions and countries monitored by PingER are shown in Fig. 2, together with the monitoring and archive sites.

Figure 2: Major regions of the world for PingER aggregation by regions

The major regions (number of countries in parentheses) are: Anglo America (2), Latin America (14), Europe (24), S.E. Europe (9), Africa (27), Mid East (7), Caucasus (3), Central Asia (8), Russia includes Belarus & Ukraine (3), S. Asia (7), China (1) and Australasia (2). The numbers in parentheses are the number of countries monitored by PingER in the region.. We also subdivide regions at times to provide better granularity. The major sub-regions are obtained by separating: Central America and the Caribbean (5) from S. America (9); Israel from the Mid-East, the Baltic States from Europe. For the purposes of characterizing the Digital Divide we also aggregate Anglo America, Europe, Australasia, Japan, Taiwan and S. Korea into the Developed Countries..

To assist in interpreting the results in terms of their impact on well-known applications, we categorize the RTTs, losses etc. into quality ranges. These are shown below in Table 2.

Table 2: Quality ranges used for loss and RTT
	Excellent	Good	Acceptable	Poor	Very Poor	Bad
Loss	<0.1%	>=0.1% & & < 1%	> =1% & < 2.5%	>= 2.5% & < 5%	>= 5% & < 12%	> 12%
RTT		<62.5ms	>=62.5ms & < 125ms	>= 125ms & < 250ms	>=250ms & < 500ms	>500ms

More on the effects of packet loss and RTT can be found in the Tutorial on Internet Monitoring & PingER at SLAC [tutorial], briefly:

At losses of 4-6% or more video-conferencing becomes irritating and non-native language speakers become unable to communicate. The occurrence of long delays of 4 seconds (such as may be caused by timeouts in recovering from packet loss) or more at a frequency of 4-5% or more is also irritating for interactive activities such as telnet and X windows. Conventional wisdom among TCP researchers holds that a loss rate of 5% has a significant adverse effect on TCP performance, because it will greatly limit the size of the congestion window and hence the transfer rate, while 3% is often substantially less serious, Vern Paxson. A random loss of 2.5% will result in Voice Over Internet Protocols (VOIP) becoming slightly annoying every 30 seconds or so. A more realistic burst loss pattern will result in VOIP distortion going from not annoying to slightly annoying when the loss goes from 0 to 1%. Since TCP throughput for the standard (Reno based) TCP stack goes as 1/(sqrt(loss) [mathis]) (see M. Mathis, J. Semke, J. Mahdavi, T. Ott, "The Macroscopic Behavior of the TCP Congestion Avoidance Algorithm",Computer Communication Review, volume 27, number 3, pp. 67-82, July 1997), it is important to keep losses low for achieving high throughput.
For RTTs, studies in the late 1970s and early 1980s showed that one needs < 400ms for high productivity interactive use. VOIP requires a RTT of < 250ms or it is hard for the listener to know when to speak.

It must be understood that these quality designations apply to normal Internet use. For high performance, and thus access to data samples and effective partnership in distributed data analysis, much lower packet loss rates may be required.

Loss

Of the two metrics loss & RTT, loss is more critical since a loss of a packet will typically cause timeouts that can last for several seconds.

Figure 3: Monthly packet loss as a function of time, seen from ESnet sites to various regions of the world. The numbers in parentheses indicate the number of monitor site / remote site pairs contributing to the medians. The orange dots show 50% improvement/year.

The following general observations can be made:

The improvement in losses is about 10-50% per year.
The best networks are achieving better than 0.1% packet for most of their sites seen from ESnet.
Russia, S.E. Europe and China are several years behind Europe and Anglo America.

Another way of looking at the losses is to see how many hosts fall in the various loss quality categories defined above as a function of time. An example of such a plot is seen in Fig 4.

Figure 4: Number of hosts measured from SLAC for each quality category from February 1998 through December 2004.

It can be seen that recently most sites fall in the good quality category. The numbers at the bottom indicate the percentage of total sites that see good packet loss at the start of the year. Also the number of sites with good quality has increased from about 55% to about 77% in the 7 years displayed. The plot also shows the total number of sites monitored from SLAC at various times. The improvements are particularly encouraging since most of the new sites are added in developing regions.

Towards the end of 2001 the number started dropping as sites blocked pings due to security concerns. The rate of blocking is such that out of 214 host that were pingable in July 2003, 33 (~15%) were no longer pingable in December 2003.

The increases towards the end of 2002 and early 2003 was due to help from the Abdus Salam Institute of Theoretical Physics (ICTP). The ICTP held a Round Table meeting on Developing Country Access to On-Line Scientific Publishing: Sustainable Alternatives [ictp] in Trieste in November 2002 that included a Proposal for Real time monitoring in Africa [africa-rtmon]. Following the meeting a formal declaration was made on RECOMMDENDATIONS OF the Round Table held in Trieste to help bridge the digital divide [icfa-rec]. The PingER project is collaborating with the ICTP to develop a monitoring project aimed at better understanding and quantifying the Digital Divide. On December 4th the ICTP electronic Journal Distribution Service (eJDS) sent an email entitled Internet Monitoring of Universities and Research Centers in Developing Countries [ejds-email] to their collaborators informing them of the launch of the monitoring project and requesting participation. By January 14th 2003, with the help of ICTP, we added about 23 hosts in about 17 countries including: Bangladesh, Brazil, China, Columbia, Ghana, Guatemala, India (Hyderabad and Kerala), Indonesia, Iran, Jordan, Korea, Mexico, Moldova, Nigeria, Pakistan, Slovakia and the Ukraine.

The increase towards the end of 2003 was spurred by preparations for the second Open Round Table on Developing Countries Access to Scientific Knowledge: Quantifying the Digital Divide 23-24 November Trieste, Italy and the WSIS conference and associated activities in Geneva December 2003.

The increases in 2004 were due to adding new sites especially in Africa, S. America, Russia and several outlying islands. See Fig. 1 and section “Accomplishments since last report”.

Fig. 5 shows the world's connected population fractions obtained by dividing countries up by loss quality seen from the US, and adding the connected populations for the countries (we obtained the population/country figures from "How many Online" [nua]) for a given loss quality to get a fraction compared to the total world's connected population.

Figure 5: Fraction of the world's connected population in countries with measured loss performance in 2001 and Dec '2003

It can be seen that in 2001, <20% of the population lived in countries with good to acceptable packet loss. By December 2003 this had risen to 75%. The coverage of Pinger has also increased from about 70 countries at the start of 2003 to over 100 in January 2004. This in turn reduced the fraction of the connected population for which PingER has no measurements.

RTTs

Unfortunately there are limits to the minimum RTT due to the speed of light in fibers or electricity in copper. Typically today, the minimum RTTs for terrestrial circuits are about 2 * distance / (0.6 * (0.6 * c)), where c is the velocity of light, the factor of 2 accounts for the round-trip, 0.6*c is roughly the speed of light in fiber and the extra 0.6 allows roughly for the delays in network equipment such as routers. For geostationary satellites links the minima are between 500 and 600ms. For U.S. cross country links (e.g. from SLAC to BNL) the typical minimum RTT (i.e. a packet sees no queuing delays) is about 70 msec.

Anglo American and European sites have been improving by between 10 & 20% per year. Japan and Russia are improving at a slower rate and S. E. Europe & mainland China at a faster rate. The improvements are due to faster links (less time clocking the bits in and out of the network equipment), faster routers and improved routes.

Fig. 6, shows the RTT from U.S. to the world in January 2000 and December 2003. It also indicates which countries of the world contain sites that are monitored (in the Jan 2000 map countries in green are not monitored, in the Dec 2003 apart from the US unmonitored countries are left white).

Figure 6: Average monthly RTT measured from U.S to various countries of the world for January 2000 and August 2002. Countries shaded white were not measured.

It is seen that the number of countries with satellite links (> 600ms RTT or dark red) has decreased markedly in the 3 years shown. Today satellite links are used in places where it is hard or unprofitable to pull terrestrial-lines (typically fibers) to.

Throughput

We also combine the loss and RTT measurements using throughput = 1460Bytes[Max Transmission Unit]/(RTT * sqrt(loss)) from [mathis]. The results are shown in Fig. 7. The orange line shows a ~60% improvement/year or about a factor of 10 in 5 years.

Figure 7: Derived throughput as a function of time seen from ESnet sites to various regions of the world. The numbers in parentheses are the number of monitoring/remote host pairs contributing to the data. The lines are exponential fits to the data.

The data for several of the developing countries only extends back for a couple of years and so some care must be taken in interpreting the long term trends. With this caveat, it can be seen that the performances to regions such as S.E. Europe (in fact we have now merged S.E. Europe with the rest of Europe), Russia, and the Caucasus are catching up to the more developed countries. Latin America and China only appear to be keeping up, while India, Africa, Central Asia and the Middle East appear to be falling further behind. At the same time it is seen that the Central Asia, Russia and Latin America are 5 to 6 years behind Europe, while China and the Middle East are 7 to 8 years behind and Africa, India and the Caucasus are 8 to 9 years behind with throughputs 15 times lower than those for Europe. In fact sites in Africa and India appear to have throughputs less than that of a well connected (cable, DSL or ISDN) home in Europe or Anglo America. For more on Africa see Connectivity Mapping in Africa [ictp-jensen], African Internet Connectivity [africa] and Internet Performance to Africa [ejds-africa]).

View from Europe

To assist is developing a less N. American view of the Digital Divide, we started adding many more hosts in developing countries to the list of hosts monitored from CERN. We now have data going back for almost three and a half years that enables us to make some statements about performance as seen from Europe. Fig. 8 shows the data from CERN as of August 2004.

Figure 8: Derived throughputs to various regions as seen from CERN.

The slow increase for N. America is partially an artifact of the difficulty of accurately measuring loss with a relatively small number of pings (14,400 pings/month at 10 pings/30 minute interval, i.e. a loss of one packet ~ 1/10,000 loss rate). The very slow increase in throughput for the Middle East, is an artifact caused by initially only monitoring hosts in 2 Middle East countries (Israel and Egypt) with one (Israel) having markedly better performance (factor of 20) than anywhere else in the Middle East. As we added hosts in more Middle East Countries (starting in July 2003), the median dropped dramatically as Israel had less effect. We have added several hosts to the Mid-East based on hosts being successfully monitored from SLAC. The lines are exponential fits to the data. Apart from the special case of the Middle East mentioned above, the trends are similar to those seen from ESnet: the improvements are between 50% and 100% per year; Russia and S. E. Europe are catching up with Europe; Latin America and China are keeping up; India is falling behind. There is insufficient data at the moment to indicate how far the various regions are behind N. America or how long it will take to catch up.

Variability of performance between and within regions

The throughput results, so far presented in this report, have been measured from Anglo America or to a lesser extent from Europe. This is partially since there is more data for a longer period available for the Anglo America monitoring hosts. Table 3 shows the throughputs seen between monitoring and remote/monitored hosts in the major regions of the world. Each column is for monitoring hosts in a given region, each row is for monitored hosts in a given region. The cells are colored according to the median quality for the monitoring region/monitored region pair. White is for derived throughputs > 1000 Kbits/s (good), blue for <= 1000 Kbits/s & >500Kbits/s (acceptable), yellow for <= 500Kbits/s and > 200 Kbits/s, and magenta for <= 200Kbits/s (very poor to bad). The table is ordered by decreasing average performance. The Monitoring countries are identified by the 2 character TLD. The .ORG site is JLab. The .NET sites are APAN in Japan and the ESnet NOC at LBNL. The .GOV sites are ESnet sites (ANL, BNL, DOE-HQ, FNAL, and LANL). S. Asia is the Indian sub-continent; S.E. Asia is composed of measurements to Indonesia, Malaysia, Singapore, Thailand and Vietnam.

Table 3: Derived throughputs in kbits/s from monitoring hosts to monitored hosts by region of the world for August 2003.

There are a couple of anomalies: the Mid East measurements are almost entirely composed of measurements to Israel; the Caucasus measurements are to 2 countries with very different performance: Azerbaijan (50% loss) and Georgia (3% loss). It can be seen that in general performance is good to acceptable. Regions with very poor to bad performance include Africa, S. Asia (India), the Caucasus region and S. America. Though not broken out here, performance to S.E. Asia is generally poor.

To provide further insight into the variability in performance for various regions of the world seen from SLAC Figure 10 shows various statistical measures of the RTTs, losses and derived throughputs (the definitions of the countries comprising a regions are also shown).

Figure 9: Maximum, 75 percentile, median and 25 percentile RTTs, losses and derived throughputs for various regions measured from SLAC for August 2003.

When there are large outliers, the sites/countries with the maxima are indicated There are quite a lot of regions with outliers. Ghana (.GH) is particularly poor for Africa. The Caucasus & C. Asia are much more uniform now that the virtual Silk Road project is in place, however, the virtual Silk Road does not serve president.KZ so Kazakhstan (.KZ) stands out. For S. America, performance is improving since the AMPATH project started providing services. However, at the moment in S. America, AMPATH only reaches Argentina, Brazil, Chile and Venezuela. Sites in the other countries use a mix of commercial carriers such as Epoch (Paraguay), Savvis (Columbia), Sprint (Ecuador, Uruguay), AT&T (Peru), Level(3) (Peru). Also: even though Chile has access to AMPATH, the two sites there (PUC and UCV) use OpenTransit and Verio respectively ; the sites in Rio de Janeiro in Brzil use GEANT instead of AMPATH. Other countries that stand out as being particularly poor for their regions are Iran (.IR) and Moldova (.MD), and within Russia (.RU) the RSSI has particularly poor performance.

Africa

Africa deserves special attention, in particular since the PingER project is probably one of the few sources of performance information. Fig. 10 shows the packet losses to African countries seen from SLAC.

Figure 10: Ping packet loss to African countries seen from SLAC

IEPM-BW Results

The PingER method of measuring throughput breaks down for high speed networks due to the different nature of packet loss for ping compared to TCP, and also since PingER only measures about 14,400 pings of a given size/month between a given monitoring host/monitored host pair. Thus if the link has a loss rate of better than 1/14400 the loss measurements will be inaccurate. For a 100Byte packet, this is equivalent to a Bit Error Rate (BER) of 1 in 10⁸, and leading networks are typically better than this today (Jan 2005). For example if the loss probability is < 1/14400 then we take the loss as being 0.5 packet to avoid a division by zero, so if the average RTT for ESnet is 50msec then the maximum throughput we can use PingER data to predict is ~ 1460Bytes*8bits/(0.050sec*sqrt(0.5/14400)) or ~ 40Mbits/s and for an RTT of 200ms this reduces to 10Mbits/s.

To address this challenge and to understand and provide monitoring of high performance throughput between major sites of interest to HEP and the Grid, we developed the IEPM-BW monitoring infrastructure and toolkit. There are about 10 monitoring hosts and about 50 monitored hosts in 9 countries (CA, CH, CZ, FR, IT, JP, NL, UK, US). Both application (file copy and file transfer) and TCP throughputs are measured.

These measurements indicate that throughputs of several hundreds of Mbits/s are regularly achievable on today's production academic and research networks, using common off the shelf hardware, standard network drivers, TCP stacks etc., standard packet sizes etc. Achieving these levels of throughput requires care in choosing the right configuration parameters. These include large TCP buffers and windows, multiple parallel streams, sufficiently powerful cpus (typically better than 1 MHz/Mbit/s), fast enough interfaces and busses, and a fast enough link (> 100Mbits/s) to the Internet. In addition for file operations one needs well designed/configured disk and file sub-systems.

Though not strictly monitoring, there is currently much activity in understanding and improving the TCP stacks (e.g. [floyd], [low], [ravot]). In particular with high speed links (> 500Mbits/s) and long RTTs (e.g. trans-continental or trans-oceanic) today's standard TCP stacks respond poorly to congestion (back off too quickly and recover too slowly). To partially overcome this one can use multiple streams or in a few special cases large (>> 1500Bytes) packets. In addition new applications ([bbcp], [bbftp], [gridftp]) are being developed to allow use of larger windows and multiple streams as well as non TCP strategies ([tsnami], [udt]). Also there is work to understand how to improve the operating system configurations [os] to improve the throughput performance. As it becomes increasingly possible to utilize more of the available bandwidth, more attention will need to be paid to fairness and the impact on other users (see for example [coccetti] and [bullot]). Besides ensuring the fairness of TCP itself, we may need to deploy and use quality of service techniques such as QBSS [qbss] or TCP stacks that back-off prematurely hence enabling others to utilize the available bandwidth better [kuzmanovic]. These subjects will be covered in more detail in the companion ICFA-SCIC Advanced Technologies Report. We note here that monitoring infrastructures such as IEPM-BW can be effectively used to measure and compare the performance of TCP stacks, measurement tools, applications and sub-components such as disk and file systems and operating systems in a real world environment.

New Monitoring and Diagnostic Efforts in HEP

PingER and IEPM-BW are excellent systems for monitoring the general health and capability of the existing networks used worldwide in HEP. However, we need additional end-to-end tools to provide individuals with the capability to quantify their network connectivity along specific paths in the network and also easier to use top level navigation/drill-down tools. The former are needed to both ascertain the user's current network capability as well as to identify limitations which may be impeding the user’s ultimate (expected) network performance. The latter are needed to simplify finding the relevant data.

Most HEP users are not a "network wizard" and don't wish to become one. In fact as pointed out by Mathis and illustrated in Fig. 11, the gap in throughput between what a network wizard and a typical user can achieve is growing.

Figure 11: Bandwidth achievable by a network wizard and a typical user as a function of time. Also shown are some recent network throughput achievements in the HEP community.

Because of HEP's critical dependence upon networks to enable their global collaborations and grid computing environments, it is extremely important that more user specific tools be developed to support these physicists.

Efforts are underway in the HENP community, in conjunction with the Internet2 End-to-End (E2E) Performance Initiative [E2Epi], to develop and deploy a network measurement and diagnostic infrastructure which includes end hosts as test points along end-to-end paths in the network. The E2E piPEs project [PiPES], the NLANR/DAST Advisor project [Advisor] and the EMA (End-host Monitoring Agent) [EMA] are all working together to help develop an infrastructure capable of making on demand or scheduled measurements along specific network paths and storing test results and host details for future reference in a common data architecture. The information format will utilize the GGF NMWG [NMWG] schema to provide portability for the results. This information could be immediately used to identify common problems and provide solutions as well as to acquire a body of results useful for baselining various combinations of hardware, firmware and software to define expectations for end users.

A primary goal is to provide as "lightweight" a client component as possible to enable widespread deployment of such a system. The EMA Java Web Start client is one example of such a client, and another is the Network Diagnostic Tester (NDT) tool [NDT]. By using Java and Java Web Start, the most current testing client can be provided to end users as easily as opening a web page. The current version supports both Linux and Windows clients.

Details of how the data is collected, stored, discovered and queried are being worked out. A demonstration of a preliminary system is being shown at the Internet2 Joint-techs meeting in Hawaii on January 25th, 2004.

The goal of easier to use top level drill down navigation to the measurement data is being tackled by MonALISA [MonALISA] in collaboration with the IEPM project.

A long term goal is to merge Pinger and IEPM-BW results into a larger distributed database architecture for use by grid scheduling and network diagnostic systems. By combining general network health and performance measurement with specific end-to-end path measurements we can enable a much more robust, performant infrastructure to support HEP worldwide and help bridge the Digital Divide.

Comparison with HEP Needs

Recent studies of HEP needs, for example the TAN Report (http://gate.hep.anl.gov/lprice/TAN/Report/TAN-report-final.doc) have focused on communications between developed regions such as Europe and Anglo America. In such reports packet loss less than 1%, vital for unimpeded interactive log-in, is assumed and attention is focused on bandwidth needs and the impact of low, but non-zero, packet loss on the ability to exploit high-bandwidth links. The PingER results show clearly that much of the world suffers packet loss impeding even very basic participation in HEP experiments and points to the need for urgent action.

The PingER throughput predictions based on the Mathis formula assume that throughput is mainly limited by packet loss. The 60% per year growth curve in figure 8 is somewhat lower than the 79% per year growth in future needs that can be inferred from the tables in the TAN Report. True throughput measurements have not been in place for long enough to measure a growth trend. Nevertheless, the throughput measurements, and the trends in predicted throughput, indicate that current attention to HEP needs between developed regions could result in needs being met. In contrast, the measurements indicate that the throughput to less developed regions is likely to continue to be well below that needed for full participation in future experiments.

Comparisons with Economic and Development Indicators

Various economic indicators have been developed by the U.N. and the International Telecommunications Union (ITU). It is interesting to see how well the PingER performance indicators correlate with the economic indicators. The comparisons are particularly interesting in cases where the agreement is poor, and may point to some interesting anomalies or suspect data.

The Human Development Index (HDI) is a summary measure of human development (see http://hdr.undp.org/reports/global/2002/en/ ). It measures the average achievements in a country in three basic dimensions of human development:

A long and healthy life, as measured by life expectancy at birth
Knowledge, as measured by the adult literacy rate (with two-thirds weight) and the combined primary, secondary and tertiary gross enrolment ratio (with one-third weight)
A decent standard of living, as measured by GDP per capita (PPP US$).

Figure 12: Comparisons of PingER losses seen from N. America to various countries versus various U.N. Development Programme (UNDP) indicators.

The Network Readiness Index (NRI) from the Center for International Development, Harvard University (see http://www.cid.harvard.edu/cr/pdf/gitrr2002_ch02.pdf ) is a major international assessment of countries’ capacity to exploit the opportunities offered by Information and Communications Technologies (ICTs), i.e. a community’s potential to participate in the Networked World of the future. The goal is to construct a network use component that measures the extent of current network connectivity, and an enabling factors component that measures a country’s capacity to exploit existing networks and create new ones. Network use is defined by 5 variables related to the quantity and quality of ICT use. Enabling factors are based on Network access, network policy, networked society and the networked economy.

Figure 13: PingER throughputs measured from N. America vs. the Network Readiness Index.

Some of the outlying countries are identified by name. Countries at the bottom right of the right hand graph may be concentrating on Internet access for all, while countries in the upper right may be focusing on excellent academic & research networks.

The Digital Access Index (DAI) created by the ITU combines eight variables, covering five areas, to provide an overall country score. The areas are availability of infrastructure, affordability of access, educational level, quality of ICT services, and Internet usage. The results of the Index point to potential stumbling blocks in ICT adoption and can help countries identify their relative strengths and weaknesses.

Figure 14: PingER derived throughput vs. the ITU Digital Access Index for PingER countries monitored from the U.S.

Since the PingER Derived Throughput is linearly proportional to RTT, countries close to the U.S. (i.e. the U.S., Canada and Mexico) may be expected to have elevated Derived Throughputs compared to their DAI. We thus do not use the U.S. and Canada in the correlation fit, and they are also off-scale in Figure 13. Mexico is included in the fit, however it is also seen to have an elevated Derived Throughput. Less easy to explain is India's elevated Derived Throughput. This maybe due to the fact that we monitor university and research sites which may have much better connectivity than India in general. Belarus on the other hand apparently has poorer Derived Throughput than would be expected from its DAI. This could be an anomaly for the one host currently monitored in Belarus and thus illustrates the need to monitor multiple sites in a developing country.

The United Nations Development Programme (UNDP) introduced the Technology Achievement Index (TAI) to reflects a country's capacity to participate in the technological innovations of the network age. The TAI aims to capture how well a country is creating and diffusing technology and building a human skill base. It includes the following dimensions: Creation of technology (e.g. patents, royalty receipts); diffusion of recent innovations (Internet hosts/capita, high & medium tech exports as share of all exports); Diffusion of old innovations (log phones/capita, log of electric consumption/capita); Human skills (mean years of schooling, gross enrollment in tertiary level in science, math & engineering). Fig. 15 shows December 2003's derived throughput measured from the U.S. vs. the TAI. The correlation is seen to be positive and medium to good. The US and Canada are excluded since the losses are not accurately measureable by PingER and the RTT is small. Hosts in well connected countries such as Finland, Sweden, Japan also have their losses poorly measured by PingER and since they have long RTTs the derived throughput is likely to be low using the Mathis formula and if no packets are lost then assuming a loss of 0.5 packets in the 14,400 sent to a host in a month.

Figure 15: PingER derived throughputs vs. the UNDP Technology Achievement Index (TAI)

Accomplishments since last report

We have extended the measurements to cover more developing countries and to increase the number of hosts monitored in each developing country. As a result the number of sites monitored from SLAC has increased by about 20% (see Fig. 4), and the countries monitored has increased by about 10% to 114; several remote sites have been added in Russia (thanks to the GLORIAD collaboration); Australian sites have been unblocked for pings since early 2004, coverage of Africa has extended to cover Angola, Botswana, Eritrea, Kenya, Niger and Tanzania, we now monitor 26 (50%) of the African countries; we have also added remote sites in Bolivia, Costa Rica the Seychelles and Thailand.. In addition monitoring sites have been added in Pakistan, Brazil and India. The measurements from these sites should assist in providing a better understanding of performance within and between developing countries/regions, and from developing regions to developed regions. We have also added a monitoring site at Florida International University which provides better coverage for AMPATH and Latin America.

The collaboration with the ICTP was very fruitful to bring in contacts from developing nations with scientific interests. However, the funding has terminated and despite efforts (proposals to the EU and others) further funding is not forthcoming.

The collaboration between SLAC and the NIIT in Rawalpindi Pakistan was funded by the Pakistan Ministry Of Science and Technology and the US Department of State for one year starting September 2004. The funding is for travel only. The collaboration is successfully working on designing, building and populating a new PingER configuration database to keep track of location (city, country, region, latitude/longitude), contacts, site name, affinity groups etc. This data is already being used to provide online maps such as Fig. 1 Work is also proceeding on automating the process of generating graphs of performance aggregated by region (e.g. Figs. 3, 4, 8, 9).

We still spend much time working with contacts to unblock pings to their sites (for example ~15% of hosts pingable in July 2003 were no longer pingable in December 2003). It is unclear how cost-effective this activity is. It can take many emails to explain the situation, sometimes requiring restarting when the problem is passed to a more technically knowledgeable person. Even then there are several unsuccessful cases where even after many months of emails and the best of intentions the pings remain blocked. One specific cases are for all university sites in Vietnam. We were successful in getting Australian sites unblocked earlier in 2004.

Even finding knowledgeable contacts, explaining what is needed and following up to see if the recommended hosts are pingable, is quite labor intensive. More recently we have had more success by using Google to search for university web sites in specific TLDs. The downside is that this way we do not have any contacts with specific people with whom we can deal in case of problems.

We now provide online interactive access to data and reports going back to January 1988. Over the New Year holiday season we had a disk fail in the RAID array holding the PingER data at SLAC. This was followed by a second disk failing during the reconstruction. Attempts to recover the data from the RAID array were eventually unsuccessful. As part of the attempts at recovery we also succeeded in re-constructing most of the data from the PingER archive at FNAL. The FNAL data is recorded in a different format so if we had to use it, then results from certain seldom used metrics would have been lost.

Presentations:

Modeling Global Internet Dynamics, presented by Robert Baker, University of New England, September 2004.
PingER Project, presented by Les Cottrell at the DoE 2004 PI Network Research Meeting, FNAL Sept Sep 15-17 '04.
ICFA/SCIC Network Monitoring, August 2004, prepared by Les Cottrell for ICFA meeting, August 2004.
PingER performance to Bangladesh prepared by Les cottrell, presented by Prof. Hilda Cerdeira in Bangladesh, May 2004.
PingER Methodology, Uses and Results, presented by Les Cottrell at the Extending the Reach of Advanced Networking: Special International Workshop, Arlington, VA., April 22, 2004.
End-to-end Internet Performance Today presented by Les Cottrell at the Extending the Reach of Advanced Networking: Special International Workshop, Arlington VA, April 22, 2004.
WAN Monitoring, presented by Les Cottrell at the Joint Engineering Task Force Roadmap Workshop, JLab April 13-15, 2004. See also the 2 page written summary.
ICFA/SCIC Network Monitoring, presented by Les Cottrell at the Digital Divide and HEPGrid Workshop, UERJ Rio de Janeiro, Brazil, Feb 16 20, 2004.
SLAC IEPM PingER and BW monitoring & tools, presented by Les Cottrell at LBL, Jan 2004.
Measurements of Internet performance for NIIT, Pakistan Jan 2004, prepared by Les Cottrell, Jan 2004

Summary

Internet performance is improving each year with losses typically improving by 40-50% per year and RTTs by 10-20% and, for some regions such as S. E. Europe, even more. Geosynchronous satellite connections are still important to countries with poor telecommunications infrastructure or in remote land-locked regions. In general for HEP countries satellite links are being replaced with land-line links with improved performance (in particular for RTT).

Links between the more developed regions including Anglo America, Japan and Europe are much better than elsewhere (5 - 10 times more throughput achievable). Regions such as S. E. Europe, Russia and Latin America appear to be catching up with developed regions such as Europe. Latin America and China only appear to be keeping up, while India, Africa, Central Asia and the Middle East appear to be falling further behind. At the same time the Central Asia, Russia and Latin America are 5 to 6 years behind Europe, while China and the Middle East are 7 to 8 years behind and Africa, India and the Caucasus are 8 to 9 years behind with throughputs 15 times lower than those for Europe. In fact sites in Africa and India appear to have throughputs less that that of a well connected (cable, DSL or ISDN) home in Europe or Anglo America. For more on Africa see Connectivity Mapping in Africa [ictp-jensen], African Internet Connectivity [africa] and Internet Performance to Africa [ejds-africa]). Though there is less extensive data, similar results are seen in measurements made from Europe. Countries/regions with particularly bad connections include the Caucasus, India, and Africa. There has been a dramatic improvement in the Internet performance for most of the world's connected population in the last 3 years.

There is a positive correlation between the various economic and development indices. Besides being useful in their own right these indices are an excellent way to illustrate anomalies and for pointing out measurement/analysis problems. The large variations between sites within a given country illustrate the need for careful checking of the results and the need for multiple sites/country to identify anomalies. The ICFA-SCIC "Digital Divide" report will dwell in more detail on many of the issues of the performance differences for the developed and less well-developed countries.

Recommendations

There is interest from ICFA, ICTP and others to extend the monitoring further to countries with no formal HEP programs, but where there are needs to understand the Internet connectivity performance in order to aid the development of science. Africa is a region with many such countries.

Extend the monitoring from within developing countries to provide performance within developing regions, between developing regions and from developing regions to developed regions..

We should ensure there are >=2 remote sites monitored in each Developing Country. All results should continue to be made available publicly via the web, and publicized to the HEP community and others. Typically HENP leads other sciences in its needs and developing an understanding and solutions. The outreach from HENP to other sciences is to be encouraged. The results should continue to be publicized widely.

We need assistance from ICFA and others to find sites to monitor and contacts in the following countries:

Latin America: Honduras, Belize, Panama
Vietnam*
Belarus^h (need > 1)
Africa: Burkino Faso, Egypt, Ghana, Malawi, Nigeria, Senegal, Somalia, Sudan (need > 1/country), Libya, (have none)

Depending on availability of funding:

simplify and where possible automate the procedures to analyze and create the summary statistical information (graphs and tables seen in the current report) at regular intervals;
develop automated methods to discover non-responsive hosts, make extra tests to pin-point reasons for non-responsiveness, and report to administrator together with contact email addresses.

Although not a recommendation per se, it would be disingenous to finish without noting the following. SLAC & FNAL are the leaders in the PingER project. The funding for the PingER effort came from the DoE MICS office since 1997, however it terminated at the end of the September 2003, since it was being funded as research and the development is no longer regarded as a research project. To continue the effort at a minimum level (maintain data collection, explain needs, reopen connections, open firewall blocks, find replacement hosts, make limited special analyses, prepare & make presentations, respond to questions) would probably require central funding at a level of about 50% of a Full Time Equivalent (FTE) person, plus travel. To extend the and enhance the project, fix known non-critical bugs, improve visualization, automate reports generated by hand today, find new country site contacts, add route histories and visualization, automate alarms, update web site for better navigation, add more Developing Country monitoring sites/countries, improve code portability) interestingly is currently being addressed by the MAGGIE-NS project with NIIT in Pakistan funded for one year by the US Department of State and the Pakistani Ministry Of Science and Technology (MOST). Without funding, for the operational side, the future of PingER and reports such as this one is unclear, and the level of effort sustained in 2003 and 2004 will not be possible in 2005. Many agencies/organizations have expressed interest (e.g DoE, ESnet, NSF, ICFA, ICTP, IDRC, UNESCO) in this work but none can (or are allowed to) fund it..

Appendix: Countries in PingER Database

The following table lists the 115 countries currently (January 1st 2005) in the PingER database. Such countries contain zero (the Vietnam hosts we used to monitor now block pings, and we are unable to find a host that does not block pings) or more sites that are being or have been monitored by PingER from SLAC. The number in the column to the right of the country name is the number of hosts monitored in that country. The number cell is colored red for zero hosts, yellow for one host for the country and green for 2 or more hosts for the country. The 37 countries marked in orange are developing countries for which we only monitor one site in the country.

References

[Advisor] http://dast.nlanr.net/Projects/Advisor/
[africa] Mike Jensen, "African Internet Connectivity". Available http://www3.sn.apc.org/africa/afrmain.htm
[africa-rtm] Enrique Canessa, "Real time network monitoring in Africa - A proposal - (Quantifying the Digital; Divide)". Available http://www.ictp.trieste.it/~ejds/seminars2002/Enrique_Canessa/index.html
[bbcp] Andrew Hanushevsky, Artem Trunov, and Les Cottrell, "P2P Data Copy Program bbcp", CHEP01, Beijing 2002. Available at http://www.slac.stanford.edu/~abh/CHEP2001/p2p_bbcp.htm
[bbftp] "Bbftp". Available http://doc.in2p3.fr/bbftp/.
[bullot] "TCP Stacks Testbed", Hadrien Bullot and R. Les Cottrell. Avialble at http://www-iepm.slac.stanford.edu/bw/tcp-eval/
[coccetti] "TCP STacks on Production Links", Fabrizzio Coccetti and R. Les Cottrell. Available at http://www-iepm.slac.stanford.edu/monitoring/bulk/tcpstacks/
[E2Epi] http://e2epi.internet2.edu/
[ejds-email] Hilda Cerdeira and the eJDS Team, ICTP/TWAS Donation Programme, "Internet Monitoring of Universities and Research Centers in Developing Countries". Available http://www.slac.stanford.edu/xorg/icfa/icfa-net-paper-dec02/ejds-email.txt
[ejds-africa] "Internet Performance to Africa" R. Les Cottrell and Enrique Canessa, Developing Countries Access to Scientific Knowledge: Quantifying the Digital Divide, ICTP Trieste, October 2003; also SLAC-PUB-10188. Available http://www.ejds.org/meeting2003/ictp/papers/Cottrell-Canessa.pdf
[ejds-pinger] "PingER History and Methodology", R. Les Cottrell, Connie Logg and Jerrod Williams. Developing Countries Access to Scientific Knowledge: Quantifying the Digital Divide, ICTP Trieste, October 2003; also SLAC-PUB-10187. Available http://www.ejds.org/meeting2003/ictp/papers/Cottrell-Logg.pdf
[EMA] http://monalisa.cern.ch/EMA/
[floyd] S. Floyd, "HighSpeed TCP for Large Congestion Windows", Internet draft draft-floyd-tcp-highspeed-01.txt, work in progress, 2002. Available http://www.icir.org/floyd/hstcp.html
[gridftp] "The GridFTP Protocol Protocol and Software". Available http://www.globus.org/datagrid/gridftp.html
[host-req] "Requirements for WAN Hosts being Monitored", Les Cottrell and Tom Glanzman. Available at http://www.slac.stanford.edu/comp/net/wan-req.html
[icfa-98] "May 1998 Report of the ICFA NTF Monitoring Working Group". Available http://www.slac.stanford.edu/xorg/icfa/ntf/
[icfa-mar02] "ICFA/SCIC meeting at CERN in March 2002". Available http://www.slac.stanford.edu/grp/scs/trip/cottrell-icfa-mar02.html
[icfa-jan03] "January 2003 Report of the ICFA-SCIC Monitoring Working Group". Available http://www.slac.stanford.edu/xorg/icfa/icfa-net-paper-dec02/

[icfa-jan04] "January 2004 Report of the ICFA-SCIC Monitoring Working Group". Available http://www.slac.stanford.edu/xorg/icfa/icfa-net-paper-jan04/
[iepm] "Internet End-to-end Performance Monitoring - Bandwidth to the World Project". Available http://www-iepm.slac.stanford.edu/bw
[ictp] Developing Country Access to On-Line Scientific Publishing: Sustainable Alternatives, Round Table meeting held at ICTP Trieste, Oct 2002. Available http://www.ictp.trieste.it/~ejds/seminars2002/program.html
[ictp-jensen] Mike Jensen, "Connectivity Mapping in Africa", presentation at the ICTP Round Table on Developing Country Access to On-Line Scientific Publishing: Sustainable Alternatives at ITCP, Trieste, October 2002. Available http://www.ictp.trieste.it/~ejds/seminars2002/Mike_Jensen/jensen-full.ppt
[ictp-rec] RECOMMDENDATIONS OF the Round Table held in Trieste to help bridge the digital divide. Available http://www.ictp.trieste.it/ejournals/meeting2002/Recommen_Trieste.pdf
[kuzmanovic] "HSTCP-LP: A Protocol for Low-Priority Bulk Data Transfer in High-Speed High-RTT Networks", Alexander Kuzmanovic, Edward Knightly and R. Les Cottrell. Available at http://dsd.lbl.gov/DIDC/PFLDnet2004/papers/Kuzmanovic.pdf
[low] S. Low, "Duality model of TCP/AQM + Stabilized Vegas". Available http://netlab.caltech.edu/FAST/meetings/2002july/fast020702.ppt
[mathis] M. Mathis, J. Semke, J. Mahdavi, T. Ott, "The Macroscopic Behavior of the TCP Congestion Avoidance Algorithm",Computer Communication Review, volume 27, number 3, pp. 67-82, July 1997
[MonALISA] http://monalisa.cacr.caltech.edu/
[NDT] http://miranda.ctd.anl.gov:7123/
[NMWG] http://www-didc.lbl.gov/NMWG/
[nua] NUA Internet Surveys, "How many Online". Available http://www.nua.ie/surveys/how_many_online/
[os] "TCP Tuning Guide". Available http://www-didc.lbl.gov/TCP-tuning/
[pinger] "PingER". Available http://www-iepm.slac.stanford.edu/pinger/; W. Matthews and R. L. Cottrell, "The PingER Project: Active Internet Performance Monitoring for the HENP Community", IEEE Communications Magazine Vol. 38 No. 5 pp 130-136, May 2002.
[pinger-deploy] "PingER Deployment". Available http://www.slac.stanford.edu/comp/net/wan-mon/deploy.html
[PiPES] http://e2epi.internet2.edu/
[qbss] "SLAC QBSS Measurements". Available http://www-iepm.slac.stanford.edu/monitoring/qbss/measure.html
[ravot] J. P. Martin-Flatin and S. Ravot, "TCP Congestion Control in Fast Long-Distance Networks", Technical Report CALT-68-2398, California Institute of Technology, July 2002. Available http://netlab.caltech.edu/FAST/publications/caltech-tr-68-2398.pdf
[tsunami] "Tsunami". Available http://ncne.nlanr.net/training/techs/2002/0728/presentations/pptfiles/200207-wallace1.ppt
[tutorial] R. L. Cottrell, "Tutorial on Internet Monitoring & PingER at SLAC". Available http://www.slac.stanford.edu/comp/net/wan-mon/tutorial.html
[udt] Y Gu, R. L Grossman, “UDT: An Application Level Transport Protocol for Grid Computing”, submitted to the Second International Workshop on Protocols for Fast Long-Distance Networks.

[un] "United Nations Population Division World Population Prospects Population database". Available http://esa.un.org/unpp/definition.html

1. In special cases, there is an option to reduce the network impact to ~ 10bits/s per monitor-remote host pair.
2. Since North America officially includes Mexico, we follow the Encyclopedia Britannica recommendation and use the terminology Anglo America (US + Canada) and Latin America. Unfortunately many of the figures use the term N. America for what should be Anglo America.
h. These countries appear in the Particle Data Group diary and so would appear to have HENP programs.
*. These countries are no longer monitored, usually the host no longer exists, or pings are blocked.