openHPI: Evolution of a MOOC Platform from LMS to SOA

Michael Totschnig, Christian Willems and Christoph Meinel

Hasso Plattner Institute, University of Potsdam, Potsdam, Germany

Keywords:

Massive Open Online Course, Learning Management System, Service Oriented Architecture.

Abstract:

This article presents a new platform for Massive Open Online Courses (MOOC), developed at the Hasso Plat-

tner Institute in Germany. After describing the evolution of the MOOC concept and the format, we explain how

we deﬁned the requirements for the platform, how we evaluated different open source learning management

systems as candidate solutions, and how the actual platform was built. The results of two ﬁrst courses deliv-

ered through the platform are presented and an outlook is given towards a planned redesign of the platform

based on a Service Oriented Software (SOA) approach.

1 MOOC FORMAT AND

HISTORY

The concept of Massive Open Online Courses

(MOOC) has been coined in 2008 in the context

of a course experiment conducted by Canadian ed-

ucational researchers Stephen Downes and George

Siemens, who opened a for credit course titled Con-

ncectivism and Connective Knowledge at the Univer-

sity of Manitoba, Canada to open registration. The

gist of the experiment consisted in encouraging learn-

ers to take the course content not as the end, but

as the beginning of an autonomous and active jour-

ney deﬁned by the connections the learner creates be-

tween resources and with co-learners (McAuley et al.,

2010).

The same concept has recently been invested with

a considerably different meaning when applied to

University courses that are migrated from the setting

of a closed physical classroom with a limited audi-

ence to an open online environment with a massive

audience. MOOCs in this second sense have origi-

nated at Stanford and MIT and some of them attracted

more than 100,000 participants.

Most discussions about MOOCs distinguish be-

tween these two concepts and formats, often referring

to them as cMOOC and xMOOC (Siemens, 2012).

At the heart of this distinction lies the tension be-

tween two opposing forces that are acting on higher

education: On the one hand, higher education in-

creasingly leaves the conﬁnes of national boundaries,

and relates to supranational frameworks (e.g. the Eu-

ropoean Bologna process) and global markets. On

the other hand, pedagogical theories that conceive the

learning process as active and social instead of pas-

sive and individual no longer are the privilege of al-

ternative and marginal institutions or educators, but

are now the backbone of commonly accepted learning

designs like collaborative or problem-based learning.

MOOCs, both the c and x variants, can be clearly sit-

uated in both movements:

• By offering a MOOC today, universities try to po-

sition themselves as global leaders of innovation,

and as educational institutions capable of deliver-

ing high-quality education on a global scale. And

in the same time invest in their attractiveness for

the most talented prospective candidates.

• While a MOOC can be organized according to

an established model of instruction, where teach-

ers transmit a well-deﬁned body of knowledge to

unknowing learners that prove their progress by

repeating back that same knowledge, it also of-

fers the opportunity of integrating collaboration

and exploration into the learning design. And

even if xMOOCs have been criticized for a lack

of pedagogical innovation, it can be hypothesized

that their massive success is due both to the effer-

vescent nature of their discussion forums and to

the learning tools available in many MOOC plat-

forms that allow creative exploration of the do-

main through virtual laboratories.

593

Totschnig M., Willems C. and Meinel C..

openHPI: Evolution of a MOOC Platform from LMS to SOA.

DOI: 10.5220/0004416905930598

In Proceedings of the 5th International Conference on Computer Supported Education (CSEDU-2013), pages 593-598

ISBN: 978-989-8565-53-2

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

2 openHPI’s COURSE FORMAT

AND REQUIREMENTS

openHPI is a platform for xMOOCs, hosted at the

Hasso Plattner Institute in Potsdam (HPI), Germany.

openHPI is the result of tele-TASK, a research and de-

velopment project conducted since 2004, which has

brought into existence an advanced lecture recording

system (Schillings and Meinel, 2002), and an online

portal

for the distribution of lecture videos. While

the tele-TASK portal has been augmented with so-

phisticated semantic web search capabilities (Sack

et al., 2009) and social web functionalities, it mainly

stayed focused on delivering lecture content to HPI’s

students allowing them to replay or to replace the

class lecture. In the advent of the MOOC format, we

see the opportunity to open up the pedagogical qual-

ity and domain expertise formerly reserved to our stu-

dents to a broader audience.

While the inspiration for our project stemmed

from the success of the Stanford courses about artiﬁ-

cial intelligence

and databases

and the MIT course

on Circuits & Electronics

, we set our ambition to de-

ﬁne a course format following a unique educational

scenario: The subject domain is split up into six

weekly units. For each week, video lectures, reading

materials, and quizzes are produced and presented in

a learning sequence. Discussion forums are set up for

each week, and actively moderated by the teaching

team. Learning progress is assessed through self-tests

that can be taken an indeﬁnite number of times, and

homework, where points are granted and collected for

the ﬁnal score, required for obtaining the certiﬁcate.

For the technical implementation of the platform,

it was clear that we did not want to rely on an ex-

ternal SAAS hosting solution, but create a platform

we could freely adapt and evolve. Teaching at HPI

focuses on IT-Systems Engineering, and the ambi-

tious project of creating a platform for thousands of

learners constitutes a very interesting challenge for

our own teaching and research. Critical success fac-

tors for the fulﬁlment of this ambition were identiﬁed

with respect to the delivery of content, the learning

process, and community building.

2.1 Learning Content

The teaching team should be empowered to concen-

trate on the quality of the content by being provided

http://www.tele-task.de/

https://www.ai-class.com/

http://www.db-class.org/

https://6002x.mitx.mit.edu/

intuitive and powerful tools for content editing and

structuring.

The presentation of the learning content should

suggest a meaningful path to novice learners while

giving advanced learners the freedom to jump to top-

ics most relevant to them.

xMOOCs draw on the distinctive engagement

qualities of video lectures, chunked into small-sized

segments. The platform must allow to embed video

content, and to enrich it with textual explanation.

openHPI uses videos from the tele-TASK portal,

where lecture video recordings already exist in form

of chunk podcasts and additional metadata extracted

from the videos.

Learning content needs to be presented in its hy-

pertextual structure, in order to allow learners to grasp

more than a linear sequence of content, i.e. the rich

connections that exist between knowledge inside and

across learning domains.

2.2 Learning Process

The learning environment must support the learning

process by allowing learners to test new competences

and by confronting them with graphical representa-

tions of their progress. Assessment tools need to be

user-friendly and interactive in a way to engage and

motivate learners. The synthetic representation of

learning progress must be easily accessible from any

part of the platform.

Learners should be able to annotate content with

personal notes only available to them and with shared

notes and comments that trigger reactions from the

teaching team and discussions in the learning com-

munity.

Learners should also be allowed to connect the

learning experience with their own tools and devices,

e.g ofﬂine consumption of lecture videos, or integra-

tion of course schedules with personal productivity

environments.

2.3 Learning Community

The distinguishing feature of the MOOC format is its

social event character: In the past, universities have

made course materials available on institutional web-

sites, and eventually provided feedback forms or dis-

cussion possibilities, for example MIT’s openCourse-

Ware project (Lerman et al., 2008). MOOCs take

place during a given time period, and hence concen-

trate the otherwise dispersed participation into a co-

herent site of collective learning. Discussion forums

should allow the teaching team to trigger participa-

tions and learners to question the content.

CSEDU2013-5thInternationalConferenceonComputerSupportedEducation

594

Many learners are less comfortable when taking

part anonymously in large groups and prefer the in-

timacy of small groups, eventually deﬁned by simi-

lar characteristics like age, location, or interests. The

platform should allow learners to ﬁnd like-minded

learners, and to deﬁne protected spaces for groups to

organize a collaborative learning experience.

The platform must not lock the user into its own

conﬁnes, but allow users to connect their learning ex-

perience to their social networks.

2.4 Quality Attributes

These functional requirements are mainly compatible

with what many learning management systems, pro-

prietary or open source, provide. But while weight-

ing quality attributes (Bass et al., 2005), like stability,

scalability, usability, look & feel, extensibility, main-

tainability, security and performance, important ques-

tions arose:

On the one hand with respect to performance:

Learning management systems have been conceived

for the context of schools or universities where class

sizes range from tens to at the most hundreds of stu-

dents. Delivering a MOOC to thousands of possibly

concurrent users needs a robust technical infrastruc-

ture, and a scalable architecture.

On the other hand, we envisioned a platform

that would be easily extensible in order to imple-

ment teaching methods like game-based learning,

peer teaching and evaluation, and to connect to vir-

tual laboratories.

Last but not least, we set as our goal, to attract

learners through an attractive online experience that

does not fall behind their experience with modern

Web 2.0 platforms.

3 openHPI’s TECHNICAL

INFRASTRUCTURE

With the start of the openHPI project, the ﬁrst task

was to build up an infrastructure that scales for a po-

tentially massive amount of users and is highly avail-

able, and to implement a suitable courseware for the

desired course format.

3.1 Courseware Implementation

The ﬁrst general decision to be taken was between

designing and implementing an own solution based

on a common web framework and the adaption and

customization of an existing (open source) software

project. When collecting the requirements for a

courseware platform it became obvious that the devel-

opment of a courseware platform from scratch would

take at least 6 months of work for a full time devel-

oper team which is a huge (ﬁnancial) effort for a ﬁrst

course with experimental character. For this reason,

we decided to build openHPI on top of an existing

tool, that would allow to us to gain experience with

the construction, administration and delivery of mas-

sive open online courses, and that both technically

and legally allowed and facilitated experimentation

with and modiﬁcation of the system.

While evaluating the landscape of existing plat-

forms for the delivery of online courses it became ap-

parent that several types of systems had each unique

advantages to offer:

• Content management systems with respect to the

ﬂexibility of the management and delivery of

learning content;

• Collaborative platforms with respect to the com-

munication features;

• Learning management systems with respect to the

support for quizzes and course design.

Finally the decision was made in favor of an open

source learning management system, where a num-

ber of potential candidates were evaluated. Among

these candidates, the projects Sakai CLE, Sakai OAE,

Canvas LMS, and Lernanta (actually a LMS frame-

work) were investigated and matched to the openHPI

requirements. The evaluation results are summarized

in Table 1.

Canvas was chosen because of its modern user

interface and the availability of crucial functional

components necessary for implementing openHPI’s

course design, mainly the sophisticated quiz engine

for managing practice exercises, assignments and ex-

ams, the discussion forum and the user-friendly in-

terface for creating and presenting the course module

structure.

Nevertheless, it was desired that the platform did

not look or behave like an actual LMS, since LMS

offer users much more freedom than openHPI needs

for the provision of its courses. A developer team of

4 full time working students put 5 weeks of work into

the adaption and customization of the Canvas LMS,

which resulted in a massive change of the Web user

interface as seen by student users. The main changes

were:

• complete rebrush of UI and platform navigation;

• additional content type for learning units: ﬂexible

video player;

• content navigation adopted to openHPI’s 6 week

course schedule;

openHPI:EvolutionofaMOOCPlatformfromLMStoSOA

595

Table 1: Comparison of Open Source Learning Management Systems.

Canvas LMS Sakai CLE Sakai OAE Lernanta

Course Landing

Page

Customizable,

Rich Text Editor

Static landing page

from course infor-

mation

Content Reposi-

tory allows ﬂexible

content

Tasks and Activity

Wall, not customiz-

able

Content Sequence Tool for course de-

sign (modules and

content blocks)

Simple syllabus

tool

Hierarchic organi-

zation of content

pages

Tasks can be re-

ordered

Learning Unit

Presentation

Generic blocks in-

side a module

As wiki pages See landing page A taks is deﬁned

by only one content

block

Quiz Environ-

ment

Very ﬂexible, so-

phisticated UI

Complex function-

ality, but sluggish

and overloaded UI

Only available

through integration

with Sakaii CLE

Discussion Forum Threaded discus-

sions, no voting

No voting No voting Participants can

post messages to

activity wall and

comments to tasks

Learning

Progress

Users can track vis-

ited units

Tracking of sub-

mitted tests

- -

Announcements Rich Text Editor Via external RSS

feed

- -

Maturity Closed-source ver-

sion used produc-

tively by schools

and universities

Used by major US

universities

Experimental In active develop-

ment, used produc-

tively P2PU

Scalability Ruby on Rails,

needs resources

Needs resources,

but works for more

than 100k users

No experience No experience

Useability, Look

& Feel

Modern UI Feels sluggish, old-

fashioned

Modern UI Clean and stylable

Extendability API, vendor plug-

ins possible

API Relies on extensi-

ble framework

• additional navigation bar for browsing the course

sequence;

• integration of a helpdesk;

• enhancement of the discussion forum (context-

speciﬁc discussions, tagging, search functional-

ity).

The adoption process kept on going during the

ﬁrst two courses, where the development team gath-

ered user feedback and wishes from the community to

leverage the running system.

3.2 Hardware Platform

The difﬁculty in building the infrastructure mainly

was about estimating the resources needed and de-

pended on a) the application-dependent demand for

resources and b) the number of expected partici-

pants. Due to a strict schedule for the ﬁrst course on

openHPI a temporary website for registration was set

up rapidly and went online about 4 weeks prior to the

launch of the actual courseware platform and 5 weeks

before the start of the ﬁrst course.

The demand for openHPI’s offering turned out to

be quite overwhelming: one week after the launch

of the temporary registration website, the database

counted more than 10.000 unique visits and about

3.000 registrations. At the point of the turnover to the

actual course platform, there were more than 10.000

registered users, of which 9.000 students enrolled for

the ﬁrst course, 30.000 unique visits and 110.000 page

impressions.

Although these numbers were a baseline for the

ﬁnal user count and could help when estimating

the resources, openHPI was put on a ﬂexible and

scalable infrastructure using a private cloud frame-

work (namely OpenNebula) for the operation of the

adopted Canvas LMS (see Figure 1).

The private cloud infrastructure allows ﬂexible

scalability and the provision of more computing re-

CSEDU2013-5thInternationalConferenceonComputerSupportedEducation

596

Figure 1: Overview of openHPI hardware architecture.

sources by simply adding additional physical hosts

to the cloud respectively shutting down servers for

reduced power consumption when the server load is

low

In the initial resource pool, two VM hosts were

provided, each with 64 cores and 64 GB RAM as well

as fast RAID 5 storage systems. Additional physical

servers provide the shared database and central ser-

vices like website monitoring. Incoming requests are

distributed by a loadbalancer appliance that also han-

dles the SSL sessions.

The video streaming was outsourced to the stream

hosting service Vimeo, which also scales with rising

demand and offers a reasonable cost model.

4 INSIGHTS FROM THE FIRST

COURSES

openHPI’s ﬁrst two courses have met with substantial

interest from the respective target audience: 13,126

learners registered for the ﬁrst course, from which

4,068 actively participated and 2,137 received the

graded certiﬁcate of successful completion. The sec-

ond course had 9891 registered learners, with 2726

active participants, and 1635 successful completions

with graded certiﬁcate. The students took about

100.000 (resp. 80.000 for the second course) self-test

attempts and submitted about 18.000 (14.000) solu-

tions of homework assignments. During the 6 weeks

of course runtime, the community generated about

1.600 forum postings in the ﬁrst course and more than

3.000 postings in the second course.

For the second course, we observed, that Sun-

day and Monday were the peak days concerning page

For a general evaluation of cloud based services in ed-

ucation, see (Ivanov, 2011).

access: up to 6.000 unique visitors generated up

to 60.000 page impressions. This peak behaviour

was actually expectable, since Monday was the day

when the weekly homework submissions were due

and the course content for the next week was pub-

lished. Concerning the dimension of the available

hardware platform, it never came to shortage of re-

sources that would have led to increased response

time for users even though the implemented software

platform turned out to be greedy for computing re-

sources.

The basic insight after ﬁrst experiences with the

implemented platform for openHPI is that traditional

learning management systems are actually not suit-

able for the operation of massive open online courses.

This insight was gained from two observations. The

ﬁrst one is a different demand concerning scalability:

traditional universities usually offer a large number of

courses with a relatively low number of participants

in each course (hundreds of students). In opposite to

this, openHPI offers only very few courses at a time,

with thousands of students in a course. This led to

massive performance problems with the Canvas LMS

since many standard system operations were imple-

mented without focus on performance but on main-

tainability and simply did not scale with participant

numbers larger than about 1.000 students. Addition-

ally, several UI elements (i.e. where lists of users

were involved) became unusable.

The second aspect is, that LMS focus on man-

agement of courses, students, tests, learning material,

etc. while the focus of a platform for massive open

online courses should be on social activity respec-

tively the activity of the students in general. Learn-

ing Management Systems are designed for courses

run and tutored by teachers or teaching teams, while

the openHPI courses should empower the students to

peer teach each other. The restricted suitability of the

LMS-based implementation of the openHPI platform

manifested e.g. in the non-availability of certain fea-

tures for students, that have been available for teach-

ers. I.e. a teacher could link every content item be-

longing to the course within related discussion post-

ings, while students would have to manually copy &

paste the respective links or describe the item in ques-

tion in their own words. In general, LMS lack the

support for mixed roles between course providers and

course participants.

5 openHPI’s NEXT GENERATION

Based on the evaluation of above shortcomings, cur-

rently a new generation for the openHPI platform is

openHPI:EvolutionofaMOOCPlatformfromLMStoSOA

597

planned. It will be based on a service-oriented archi-

tecture with the following design principles in mind:

• The new system should integrate with state-of-

the-art backend services, if they are available as

open source solutions and can be trusted to be

both scalable and reliable. For example we evalu-

ate the incorporation of digital asset management

systems like Fedora.

• Service-oriented system design similar to

component-based approaches strives at modu-

larity and re-usability. Well-deﬁned interfaces

allow to separate the construction of the system

into development tasks that can be executed by

independent teams in parallel.

• Making services consumable from different client

environments allows straightforward implementa-

tion of alternative client software, i.e. for mobile

devices (mobile websites and native apps for iOS,

Android and Windows 8).

• By decoupling the consumption of services in spe-

ciﬁc contexts from the provision of these services,

both can evolve independently with respect to the

affordances of their respective environment, for

example the integration of new UI libraries, or the

adoption of new communication protocols.

• A service-oriented architecture provides a sound

basis for the integration of the learning platform

into existing and future content production work-

ﬂows. For example, we plan to integrate openHPI

with the tele-TASK portal, were tools for collab-

orative work with e-lectures, such as live script-

ing and temporal (social) annotations have already

been implemented.

• Allow exposure of services to third-party tools

that provide value-added services to learners and

teachers (e.g. personal learning environments,

social networks or competence management (e-

portfolio) systems.

• In the global competitive MOOC landscape, plat-

forms need to provide unique learning tools in or-

der to distinguish themselves from the competi-

tion. Hence it is important that the system can

connect with external learning facilities, such as

virtualized computing labs or simulation environ-

ments for hands-on training.

REFERENCES

Bass, L., Clements, P., and Kazman, R. (2005). Software

architecture in practice. Addison-Wesley, Boston ;

Munich [u.a.].

Ivanov, I. (2011). Cloud computing in education - the inter-

section of challenges and opportunities. In Verbraeck,

A., Helfert, M., Cordeiro, J., and Shishkov, B., editors,

CSEDU 2011 - Proceedings of the 3rd International

Conference on Computer Supported Education, Vol-

ume 1, Noordwijkerhout, Netherlands, 6-8 May, 2011,

page 05. SciTePress.

Lerman, S., Miyagawa, S., and Margulies, A. (2008). Open-

courseware: Building a culture of sharing. In Iiyoshi,

T. and Kumar, M. S. V., editors, Opening Up Ed-

ucation: The Collective Advancement of Education

Through Open Technology, Open Content, and Open

Knowledge, pages 213–228. MIT Press MA,, USA.

McAuley, A., Stewart, B., Siemens, G., and Cormier, D.

(2010). The mooc model for digital practice. http://

www.elearnspace.org/Articles/MOOC

Final.pdf.

[Online; accessed 04-March-2013].

Sack, H., Baumann, B., Groß, A., and Meinel, C. (2009).

Linking tele-task video portal to the semantic web. In

9th International Conference on Innovative Internet

Community Systems (I2CS09), volume P-148 of Lec-

ture Notes in Informatics (LNI), pages 15–17, Jena,

Germany. Gesellschaft f

ur Informatik.

Schillings, V. and Meinel, C. (2002). tele-task – teleteach-

ing anywhere solution kit. In ACM SIGUCCS, Provi-

dence, USA. ACM Press.

Siemens, G. (2012). Moocs are really a platform.

http://www.elearnspace.org/blog/2012/07/25/moocs-

are-really-a-platform/. [Online; accessed 04-March-

2013].

CSEDU2013-5thInternationalConferenceonComputerSupportedEducation

598