RASP-IVR: A Low Cost

Interactive Voice Response System

Vikram Kamath Cannanure Timothy X Brown
Carnegie Mellon University Carnegie Mellon University
Pittsburgh, PA Kigali, Rwanda
kmarkiv@cmu.edu timxb@cmu.edu
ABSTRACT 1 INTRODUCTION
Interactive Voice Response (IVR) systems have demonstrated Interactive Voice Response (IVR) technologies1 provide a
that they provide an opportunity for a low-literate audi- platform for rural and low-income users in the developing
ence to express themselves and interact with information world to communicate in novel ways using phone calls from
technology systems. However, to achieve such goals in a their mobile phones. The voice-based interaction allows de-
development setting requires a low-cost system that allows signing for users with less literacy on a ubiquitous and fa-
participation of multiple stakeholders in the design process. miliar mobile-phone interface. Prior work on IVR systems
In this paper, we describe a low-cost IVR hardware system has been used for citizen journalism [26], localized educa-
called RASP-IVR that runs on a Raspberry PI and a local GSM tion for health [31] support groups for farmers in agricul-
modem. RASP-IVR is an open source appropriate system en- ture [28], and community-based entertainment such as a
gineered to nurture community driven solutions. We have music portal [23, 35], community radio [21], and viral en-
open-sourced our system and are working with a Rwandan tertainment [29]. Although these examples show that IVR
community based organization to test our system in the field. systems can potentially benefit community-based organiza-
tions (CBOs) working with users in low-resource settings,
CCS CONCEPTS a typical IVR system has high setup, call, maintenance, and
• Hardware → Sound-based input / output; • Applied com- staff training costs [25]. This makes it hard to invest in a
puting → Microcomputers; system in the early stages of a project creating a hurdle to
adoption. Therefore, before a full IVR investment, there is a
need for a simple and low-cost system that can be used to
KEYWORDS
get early feedback from the community as shown in Fig. 1.
Interactive Voice Response, Telephone Interface, Community
Content

ACM Reference Format:

Vikram Kamath Cannanure and Timothy X Brown. 2018. RASP-
IVR: A Low Cost Interactive Voice Response System. In 2nd African
Conference for Human Computer Interaction (AfriCHI ’18), December
3–7, 2018, Windhoek, Namibia. ACM, New York, NY, USA, 5 pages.
https://doi.org/10.1145/3283458.3283489

Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies
are not made or distributed for profit or commercial advantage and that
Figure 1: RASP-IVR is a low cost ($50) IVR system consisting
copies bear this notice and the full citation on the first page. Copyrights of a Raspberry Pi and a GSM modem.
for components of this work owned by others than the author(s) must
be honored. Abstracting with credit is permitted. To copy otherwise, or In the past, successful IVR systems have been designed
republish, to post on servers or to redistribute to lists, requires prior specific with active end-user involvement, for example, Avaaj Otalo
permission and/or a fee. Request permissions from permissions@acm.org.
used Wizard of Oz techniques [22], The Urban Sex Worker
AfriCHI ’18, December 3–7, 2018, Windhoek, Namibia
project used ethnography [30] and Healthline used inter-
© 2018 Copyright held by the owner/author(s). Publication rights licensed
to ACM. views [31]. Participatory design or the concept of getting
ACM ISBN 978-1-4503-6558-1/18/12. . . $15.00 1 Anexample is the familiar airline call center: “Press one for reservations.
https://doi.org/10.1145/3283458.3283489 Press two for check-in. . . . ”
AfriCHI ’18, December 3–7, 2018, Windhoek, Namibia Cannanure et al.

users and other stakeholders involved in the design process 3 RASP-IVR DESIGN
is a solution towards designing successful interventions in Our motivation for RASP-IVR was from our preliminary in-
developing countries [27]. However, the earlier uses of par- terviews with CBOs across the world interested in IVR tech-
ticipatory methods in IVR system design all worked at an nology. Firstly, our partners were concerned with the setup
abstract pre-implementation stage. Active user participation cost and were interested in a prototype before an investment
in working IVR prototypes can lead to more rapid and rele- in IVRs. Secondly, existing cloud systems allowed researchers
vant solutions. to prototype applications but the power dynamic [27] was
This paper describes RASP-IVR, a low-cost tool that can skewed towards researchers and away from participants. We
support early user engagement while building IVR systems. reflected on these tensions [19] and saw an opportunity to
RASP-IVR is an open source appropriate system that lever- build low-cost IVR systems. Lastly, following the vision of
ages open source software and existing devices to lower de- Brewer et al. [17] to promote open source in the development
ployment costs. RASP-IVR is designed for user involvement domain and from prior successes (such as Open Data Kit [18]
so developers, CBO staff, and other users can collaborate and and Commcare [33]) we have open-sourced our application.
learn from the local community before investing in a full The “RASP-IVR” name was chosen because it was built
solution. For small-scale deployments, RASP-IVR can also on the Raspberry Pi, a low-cost, Linux-based, single board
act as a low-cost long-term solution. computer. The OS is based on RASPBX [1] with a few cus-
tomizations and libraries to better support IVR interactions.
2 RELATED WORK
The system consists of a Raspberry Pi model 3 and a GSM
Existing IVR systems can be divided into open source, com- mobile telephone modem. The GSM modem interfaces with
mercial enterprise, and commercial cloud solutions. Among an Asterisk server. The Asterisk server can be managed using
open source systems, there are Asterisk [9] and FreeSwitch [3] FreePBX web interface to create and customize the IVR tree
which reduce the setup and airtime costs but require addi- or voice over IP (VoIP) parameters. We have added scripts
tional costs such as development and maintenance. Avaaj using open source libraries to extend RASPBX to support
Otalo [28] was deployed in Gujarat, India to create a social Voice messages, SMS, and local SIP transfers.
network for rural farmers but needs customization and de- Figure 2 shows the working of RASP-IVR. The GSM mo-
velopment work for other applications. Another alternative dem interfaces with voice calls and text messages to connect
is IVR Junction developed by Vashita et al. [36]. It allows it to an Asterisk server. The Asterisk server can connect
customization and has been deployed in various locations the GSM call to the local or global VoIP network based on
in India and Africa. It uses a local SIM card to lower calling the requirements. The VoIP calls can be routed to various
cost but has a setup cost of around $1000 that is prohibitive extension based on the clients touch tone input or answered
at the early design stages in a low-resource setting. Commer- through SIP clients like Yate[16], Zoiper among others using
cial systems like Awaaz.De [2], Gram Vaani [5] and Voto [4] a desktop or a mobile device through a local SIP account.
among others provide robust and customizable enterprise Additionally, the VoIP call can be transferred to external
solutions which also require significant early investment. services like Twilio to use their cloud-based services e.g.
Cloud-based services like Twilio [14], Torpo [13], Sinch [11], Speech to text engine or cloud storage if needed. However,
and Pilvo [12] require no upfront hardware or software but RASP-IVR is self-sufficient for participatory design or for
have service fees. They offer APIs to build applications using small-scale deployments.
various programming languages using skilled developers. In RASP-IVR is a very low-cost system and it consumes very
many cases, they require international calls to be paid for by little power. Our system can be set up for $50, i.e. $35 for a
the community users or the service providers. The systems raspberry PI and $15 for the GSM Modem. The system’s capa-
often have text-to-speech and other features designed for bility can be enhanced by connecting two or more modems
English and other common languages that do not cater to allowing for more concurrent calls and call routing. Adding
less-common local languages. So, although existing systems multiple modems may cause the modems to restart during
provide significant benefit, they involve upfront costs, ser- high power consumption actions like calls. This limitation
vice fees, or insufficient localization to support early user can be overcome using a low-cost USB hub which costs $10.
involvement in low-resource settings. The basic system costs $50 but the system can be augmented
We build on prior work to support active community- with low-cost tools to boost its capability based on context.
driven solutions in IVR. From a technical perspective, our so- For example, a 5V DC uninterruptible power supply costing
lution is a cost-effective and simpler version of the traditional $20 can be added to handle power outages.
IVR system using Asterisk [9] and a router. We improve on Developing countries are not well supported by cloud-
these solutions through our low-cost hardware based on a based services and even when services are present, they fall
Raspberry PI and a GSM modem to build the IVR systems.
RASP-IVR: A Low Cost
Interactive Voice Response System AfriCHI ’18, December 3–7, 2018, Windhoek, Namibia

Figure 2: The RASP-IVR system can interact with GSM lines Figure 3: Shows response times for RASP-IVR as time pro-
via voice calls or text messages. These interactions can be gresses for 10 calls per second, 20 concurrent calls and 100
transferred to the RaspberryPI server and can be transferred total calls. This tells us the responsiveness of the RASP-IVR
to VoIP or external lines based on the application design. for handling concurrent calls. Although the number of calls
is limited to the number of modems, the system can route
the call internally and handle complex scenarios such as
beyond the budget of local communities. RASP-IVR circum- conference calls.
vents this issue by using local SIMs which allows the use
of airtime bundles and existing SIM cards familiar to users.
experiment. This shows the system VoIP portion of the sys-
Table 1 shows the comparison of 1000 minutes of Twilio’s
tem can handle simultaneous calls with little delay in call
cloud-based usage fees versus 1000 minutes of the default
setup. Note that unlike the VoIP interface, the GSM interface
local calling costs to the RASP-IVR for a few countries in
dongle can handle only one call at a time. Additional dongles
East Africa and we show the percentage savings. Using a
can be added to build capacity, although each dongle would
plan or airtime bundle considerably lowers the calling costs
have a different phone number.
further (e.g. see Tanzania), this can be capitalized to further
increase savings. 4 DISCUSSION
We profiled a RASP-IVR instance using automated calls
using sipP. sipP [6] is a command line tool to simulate VoIP RASP-IVR opens the possibility of creating custom appli-
calls on so-called SIP servers. We set up a SIP channel on a cations that can support contextually relevant challenges
RASP-IVR instance (on an MTN E153 modem and Raspberry using its voice and text messaging platforms. The RASP-IVR
PI 3) and tested the system using a sipP script. We ran a enables features such as automatic redirecting of calls, user
simulation for 10 calls per second, 20 concurrent calls, and selected content, recording voice content from callers, text
100 total calls and measured the response rates. A test call message-based interaction, data collection, and personalized
dialed into a SIP channel and listened to a recorded message, content based on caller ID. However, we are reflective that
We found that RASP-IVR successfully handled all 100 of the IVR based systems like all technology needs to be designed
calls with no failure in 2 mins. The system handled 0.828 to magnify existing human development efforts [34].
calls/second. As the experiment progressed, the response While there are caveats discussed further below, advan-
times of calls converged towards 0.015 seconds. Figure 3 tages of IVR-based systems include the following. A purely
shows the plot of response times over the duration of the voice-based system appeals to low-literate users as compared
to text-message or web-based applications. IVR systems are
accessed through the already familiar process of calling on
Table 1: Comparison of airtime cost for RASP-IVR widely available mobile phones. The mobile phone access
compared to Twilioś (Cloud) airtime charges for 1000 can be more private and less stigmatizing than physical vis-
mins for countries in East Africa its to sensitive services such as mental health counseling or
family planning. Caller ID allows the IVR system to call back
Country Twilio RaspIVR(Local) Saving% dropped calls and to connect activity from the same phone
across different calls. Lastly, the voice signal itself contains
Kenya $106 $70[10] 34%
useful information about the speaker including estimates of
Uganda $430 $64[8] 85%
age, weight, gender, stress, and other health factors [32]].
Rwanda $405 $52[7] 87%
We have been working with a few partners in the health
Tanzania $500 $22(plan)[15] 96%
sector in Rwanda. Our partners have existing face-to-face
AfriCHI ’18, December 3–7, 2018, Windhoek, Namibia Cannanure et al.

operations and we hope to build systems to extend the cover- Future Work
age and the reach of their services [34]. IVR can reach clients We have presented RASP-IVR a low-cost open-source system
who cannot travel to existing centers providing extensions, to design IVR systems with the local communities in the
albeit limited, of their existing services. Interestingly, one developing world. We expect to extend our current design in
NGO partner periodically travels to outlying villages and a field evaluation. We are working with Rwandan community
views the IVR as a tool for driving visitors to these outreach based organizations to test our system in the field. We expect
activities. The IVR can reduce the workload of staff through this to open new directions for RASP-IVR and also help
automated services that cover outside of normal hours or col- correct existing assumptions we have in our design.
lect and analyze user data. One partner will use this feature We expect RASP-IVR to act as an IVR plugin to Open Data
to automate reporting to donor sponsors. Another partner Kit [18] and serve as an alternative to the deprecated ODK
views the IVR as part of a larger transition from paper to Voice [24]. For data management we see opportunities to
electronic health records (EHR). The IVR could be integrated integrate tools like ODK Aggregate [18] or Open MRS [20].
with the EHR system [24] and a web application where the We will be releasing RASP-IVR with sample code and
IVR collects caller data which prompts the web application detailed tutorials for people to set up IVR systems. We hope
to pull up the client EHR before speaking with a counselor. to create a web platform for community-driven RASP-IVR
The call is followed up with a text message to the client and solutions. Our work in progress portal can be found at
another message to the field office closest to the client.
Against these benefits, there are, however, some caveats. http://bit.ly/rasp-ivr.
The IVR system is language specific and prompts need to
5 CONCLUSION
be recorded in the different languages and dialects of the
user population. While the phone interface is familiar, IVR Interactive Voice Response (IVR) systems have demonstrated
systems are inhibited by the social and cultural norms of de- that they provide an opportunity for a low-literate audience
veloping communities. Users more familiar with face-to-face to express themselves and interact with information tech-
interaction may not be receptive to automated interactions nology systems. However, IVR systems have operational
and data queries. Indeed, one partner anticipates that they cost and setup overhead creating a barrier to entry. In the
will get the best response after their clients are shown how past, successful IVR systems have been designed with ac-
to use the IVR system during face-to-face meetings. We also tive end-user involvement. However, to achieve such goals
found that access to phones themselves are a limitation for in a development setting requires a low-cost system that
low-resource users. Phones are not always charged due to allows participation of multiple stakeholders in the design
cost and availability of electricity. Many times phones are process. In this paper, we have described RASP-IVR a low-
shared so that the same phone might be used by different cost and low power IVR system to help build community-
users and a given user might use different phones. Follow driven solutions for the developing world. RASP-IVR is an
up messages and calls may not reach the intended person. open source appropriate system designed for early partici-
Further, the use of a shared phone can prompt interest and patory development before transition to robust large-scale
suspicion about who is being called. Finally, the IVR sys- deployments. The cost effective nature of RASP-IVR allows
tems raise a number of privacy concerns from how collected room for closer and deeper stakeholder engagement to pilot
data is stored, transmitted, and reported to how one user is early-stage systems before investing in a full-fledged IVR
prevented from accessing another’s data. solution. We have open sourced our system and continue to
From the developer side, the current system requires pro- develop our tool through a field evaluation with a Rwandan
gramming expertise which may not be readily available for a community based organization. While focused on CBOs in
CBO. We found that not all modem dongles worked with the this paper, RASP-IVR is useful for local entrepreneurs and
system and it was trial and error to find a model that worked outside organizations (such as NGOs) to develop and test
correctly. In some cases, it was a hardware issue where the IVR solutions as well.
dongle would not communicate with the Raspberry Pi. In
others, it was a network issue: One service provider did not ACKNOWLEDGMENTS
transfer the touch tone signals in mobile-to-mobile calls (in- We would like to thank Placide Hakuzweyezu and Saphani
cluding calls to the GSM dongle). The system, at present, Bazimya from the Fablab Rwanda for sharing their modem
can handle one call at a time and is not appropriate for high- and brainstorming ideas during development. We also thank
calling volume deployments. For our partners who anticipate our research assistants Justin Bigwi and Nebiyou Yismaw
modest volumes the RASP-IVR is appropriate. If volume were who helped verify the airtime numbers. We are very grateful
to grow, they could justify a transfer to one of the more scal- to the Human Computer Interaction Institute at Carnegie
able solutions in Section 2. Mellon and the CMU Africa campus for supporting this work.
RASP-IVR: A Low Cost
Interactive Voice Response System AfriCHI ’18, December 3–7, 2018, Windhoek, Namibia

REFERENCES [23] Zahir Koradia, Piyush Aggarwal, Aaditeshwar Seth, and Gaurav Luthra.
[1] [n. d.]. Asterisk for Raspberry Pi. http://www.raspberry-asterisk.org/. 2013. Gurgaon Idol: A Singing Competition over Community Radio
(Accessed on 03/26/2018). and IVRS. In Proceedings of the 3rd ACM Symposium on Computing
[2] [n. d.]. Awaaz.De | Mobile Solutions for Social Impact. https://awaaz. for Development (ACM DEV ’13). ACM, New York, NY, USA, Article 6,
de/. (Accessed on 03/02/2018). 10 pages. https://doi.org/10.1145/2442882.2442890
[3] [n. d.]. FreeSWITCH. https://freeswitch.org/. (Accessed on [24] Adam Lerer, Molly Ward, and Saman Amarasinghe. 2010. Evaluation
03/02/2018). of IVR Data Collection UIs for Untrained Rural Users. In Proceedings
[4] [n. d.]. A global social enterprise improving lives via mobile - Viamo. of the First ACM Symposium on Computing for Development (ACM DEV
https://viamo.io/. (Accessed on 03/02/2018). ’10). ACM, New York, NY, USA, Article 2, 8 pages. https://doi.org/10.
[5] [n. d.]. gramvaani | community-powered-technology. http://www. 1145/1926180.1926183
gramvaani.org/. (Accessed on 03/02/2018). [25] Eleanor R. Marchant. 2016. Interactive voice response and radio for
[6] [n. d.]. mojolingo/sippy_cup: Create SIP load test scenarios the peacebuilding: A macro view of the literature and experience from the
easy way. https://github.com/mojolingo/sippy_cup. (Accessed on field. Technical Report. Annenberg School of Communications Univer-
07/13/2018). sity of Pennsylvania. http://www.global.asc.upenn.edu/app/uploads/
[7] [n. d.]. MTN Rwanda - Welcome. http://www.mtn.co.rw/Content/ 2016/02/IVR-Lit-Review_Final.pdf.
Pages/40/MTN_Per_Second. (Accessed on 07/13/2018). [26] Preeti Mudliar, Jonathan Donner, and William Thies. 2012. Emergent
[8] [n. d.]. MTN Uganda | Products | MTN Call Per Second. https://www. Practices Around CGNet Swara, Voice Forum for Citizen Journalism in
mtn.co.ug/en/products/tariff-plans/Pages/mtn-call-per-second.aspx. Rural India. In Proc. of the Fifth International Conference on Information
(Accessed on 07/13/2018). and Communication Technologies and Development (ICTD ’12). ACM,
[9] [n. d.]. Open Source Communications Software | Asterisk Official Site. New York, NY, USA, 159–168. https://doi.org/10.1145/2160673.2160695
https://www.asterisk.org/. (Accessed on 03/02/2018). [27] Peter Oakley. 1991. Projects with people: The practice of participation in
[10] [n. d.]. Safaricom Rates | ILRI/ICRAF ICT Portal. http://ict. rural development. International Labour Organization. 284 pages.
worldagroforestry-ilri.org/safcom. (Accessed on 07/13/2018). [28] Neil Patel, Deepti Chittamuru, Anupam Jain, Paresh Dave, and Tapan S.
[11] [n. d.]. Sinch | Voice API, Verification & Cloud VoIP made Simple. Parikh. 2010. Avaaj Otalo: A Field Study of an Interactive Voice Forum
https://www.sinch.com/. (Accessed on 03/26/2018). for Small Farmers in Rural India. In Proceedings of the SIGCHI Confer-
[12] [n. d.]. SMS API, Voice API, & Global Carrier Provider | Plivo. https: ence on Human Factors in Computing Systems (CHI ’10). ACM, New
//www.plivo.com/. (Accessed on 03/26/2018). York, NY, USA, 733–742. https://doi.org/10.1145/1753326.1753434
[13] [n. d.]. Tropo: Home. https://www.tropo.com/. (Accessed on [29] Agha Ali Raza, Mansoor Pervaiz, Christina Milo, Samia Razaq, Guy
03/26/2018). Alster, Jahanzeb Sherwani, Umar Saif, and Roni Rosenfeld. 2012.
[14] [n. d.]. Twilio - Communication APIs for SMS, Voice, Video and Viral Entertainment As a Vehicle for Disseminating Speech-based
Authentication. https://www.twilio.com/. (Accessed on 03/07/2018). Services to Low-literate Users. In Proceedings of the Fifth Interna-
[15] [n. d.]. Voice & SMS Overview. https://vodacom.co.tz/en/ tional Conference on Information and Communication Technologies
voice-and-sms-overview/. (Accessed on 07/13/2018). and Development (ICTD ’12). ACM, New York, NY, USA, 350–359.
[16] [n. d.]. Yate. http://yate.ro/. (Accessed on 07/13/2018). https://doi.org/10.1145/2160673.2160715
[17] Eric Brewer, Michael Demmer, Bowei Du, Melissa Ho, Matthew Kam, [30] Nithya Sambasivan, Julie Weber, and Edward Cutrell. 2011. Designing
Sergiu Nedevschi, Joyojeet Pal, Rabin Patra, Sonesh Surana, and Kevin a Phone Broadcasting System for Urban Sex Workers in India. In
Fall. 2005. The Case for Technology in Developing Regions. Computer Proceedings of the SIGCHI Conference on Human Factors in Computing
38, 6 (June 2005), 25–38. https://doi.org/10.1109/MC.2005.204 Systems (CHI ’11). ACM, New York, NY, USA, 267–276. https://doi.
[18] Waylon Brunette, Mitchell Sundt, Nicola Dell, Rohit Chaudhri, Nathan org/10.1145/1978942.1978980
Breit, and Gaetano Borriello. 2013. Open Data Kit 2.0: Expanding and [31] J. Sherwani, N. Ali, S. Mirza, A. Fatma, Y. Memon, M. Karim, R. Tongia,
Refining Information Services for Developing Regions. In Proceedings and R. Rosenfeld. 2007. HealthLine: Speech-based access to health
of the 14th Workshop on Mobile Computing Systems and Applications information by low-literate users. In 2007 International Conference on
(HotMobile ’13). ACM, New York, NY, USA, Article 10, 6 pages. https: Information and Communication Technologies and Development. 1–9.
//doi.org/10.1145/2444776.2444790 https://doi.org/10.1109/ICTD.2007.4937399
[19] Kim VL England. 1994. Getting personal: Reflexivity, positionality, and [32] Rita Singh, Bhiksha Raj, and Deniz Gencaga. 2016. Forensic anthropom-
feminist research. The Professional Geographer 46, 1 (1994), 80–89. etry from voice: an articulatory-phonetic approach. In Information and
[20] Suranga Nath Kasthurirathne. 2012. Designing and Developing an Communication Technology, Electronics and Microelectronics (MIPRO),
Open Source Medical Informatics Module. XRDS 19, 2 (Dec. 2012), 2016 39th International Convention on. IEEE, 1375–1380.
20–23. https://doi.org/10.1145/2382856.2382866 [33] T Svoronos, P Mjungu, R Dhadialla, R Luk, C Zue, J Jackson, and N
[21] Konstantinos Kazakos, Siddhartha Asthana, Madeline Balaam, Mona Lesh. 2010. CommCare: Automated quality improvement to strengthen
Duggal, Amey Holden, Limalemla Jamir, Nanda Kishore Kannuri, community-based health. Weston: D-Tree International (2010).
Saurabh Kumar, Amarendar Reddy Manindla, Subhashini Arcot [34] Kentaro Toyama. 2011. Technology as amplifier in international devel-
Manikam, GVS Murthy, Papreen Nahar, Peter Phillimore, Shreyaswi opment. In Proceedings of the 2011 iConference. ACM, 75–82.
Sathyanath, Pushpendra Singh, Meenu Singh, Pete Wright, Deepika [35] Aditya Vashistha, Edward Cutrell, Gaetano Borriello, and William
Yadav, and Patrick Olivier. 2016. A Real-Time IVR Platform for Com- Thies. 2015. Sangeet Swara: A Community-Moderated Voice Forum
munity Radio. In Proceedings of the 2016 CHI Conference on Human in Rural India. In Proceedings of the 33rd Annual ACM Conference on
Factors in Computing Systems (CHI ’16). ACM, New York, NY, USA, Human Factors in Computing Systems (CHI ’15). ACM, New York, NY,
343–354. https://doi.org/10.1145/2858036.2858585 USA, 417–426. https://doi.org/10.1145/2702123.2702191
[22] J. F. Kelley. 1984. An Iterative Design Methodology for User-friendly [36] Aditya Vashistha and William Thies. 2012. IVR Junction: Building
Natural Language Office Information Applications. ACM Trans. Inf. Scalable and Distributed Voice Forums in the Developing World. In Pre-
Syst. 2, 1 (Jan. 1984), 26–41. https://doi.org/10.1145/357417.357420 sented as part of the 6th USENIX/ACM Workshop on Networked Systems
for Developing Regions. USENIX, Boston, MA.