You are on page 1of 8

FOCUS: DevOps: LESSONS LEARNED

Innovative I find myself more and more ex-


asperated with the great inflexible
sets of rules that many companies

Practices for pour into concrete and sanctify


as methods. … Use the prevailing
method only as a starting point

Knowledge
for tailoring.

—Tom DeMarco (1982)

Sharing in Large- AGILE METHODS EMERGED from


the inability of conventional (for
example, waterfall) methods to de-

Scale DevOps liver software to meet the needs of


a rapidly changing environment.
Initially, agile methods were consid-
ered to have a “home ground” where
Aymeric Hemon, ESSCA School of Management and University of Nantes they were best suited; namely, small
projects with colocated developers
Brian Fitzgerald, Lero and University of Limerick in noncritical domains. The home
Barbara Lyonnet, University of Nantes ground theory has been challenged
considerably during the past 15 years
Frantz Rowe, University of Nantes and SKEMA Business School
as researchers and practitioners tai-
lored the application of agile meth-
// Agile development methods and DevOps ods through new roles, ceremonies,
require adaptation during implementation to meet and artifacts to meet the needs of the
development context; for example,
the needs of a constantly changing software- in critical and regulated domains.
development environment. The emergence of In addition, a bottleneck emerged
due to a lack of alignment between the
knowledge-sharing practices for large-scale operations function, which coordinates
DevOps has not been the subject of much the software release, and development
function. Consequently, releases to cus-
research. Our in-depth case study, consisting
tomers took more time. To solve the
of 106 interviews at a multinational company problem, Debois1 advocated for more
operating in a DevOps-at-scale environment, collaboration between the develop-
ment and operations functions through
identified a number of innovative practices. // tighter integration, a rapprochement
called DevOps. The term comes from
the fusion of two words related to spe-
cific component activities: development
and operations.
While agile methods can achieve a
more frequent cadence of software de-
velopment and better alignment with
customer expectations, DevOps strives
for a continuous delivery of value
Digital Object Identifier 10.1109/MS.2019.2958900
through constant integration, delivery,
Date of current version: 16 April 2020 and deployment. DevOps is, hence, an

30 I E E E S O F T WA R E | PUBLISHED BY THE IEEE COMPUTER SO CIE T Y 0 74 0 -74 5 9 / 2 0 © 2 0 2 0 I E E E

Authorized licensed use limited to: UNIVERSIDAD VERACRUZANA. Downloaded on February 19,2023 at 01:07:05 UTC from IEEE Xplore. Restrictions apply.
extension of agile2 to the entire soft- transfer when moving to large-scale and sharing. Three of the six KS chal-
ware-delivery pipeline, aiming to opti- DevOps. Despite the need for inno- lenges are particularly salient. First,
mize the lead time between code writing vation and tailoring, as expressed by the ability to self-organize could be
and its use by end users in a real produc- DeMarco in the opening quote, new affected by scaling up DevOps due
tion environment. KS practices for large-scale DevOps to a larger number of team members
We consider large-scale DevOps as have not been extensively studied. In covering a wider organizational pe-
we focus on the four DevOps pillars that respect, Nielsen et al. proposed rimeter. Second, notwithstanding a
(culture, automation, measurement, a DevOps KS framework9 and used higher level of automation at scale,
and sharing) that involve a large the combination, externalization, more frequent commits (releases) re-
number of actors, systems, and inter- socialization, and internalization di- quire manual coding adjustment,16
dependencies, with more than two mensions to increase awareness of KS which needs a common coordination
teams working on the same project3 modes within the organization. They mechanism and, hence, more intense
and applying continuous integration, concluded that the size of the team cross-functional collaboration. Third,
delivery, and deployment. Roles and and company influences the move scaling DevOps could impact the KS
responsibilities do still matter in the toward continuous delivery and due to more friction with the organi-
agile world, particularly when using DevOps. They showed that larger zational structure and red tape.
prescriptive methods, such as Scrum. companies need a more structured
DevOps breaks silos and, therefore, plan when moving to DevOps to en- Context
possibly blurs the lines (such as orga- sure K S a m o n g t h e i r I T t e a m s , Our study was conducted in a mul-
nizational and hierarchical) between while smaller organizations do not tinational company (with more than
jobs, roles, collaborations, responsi- need a similar strategy. Consequently, 100,000 employees) that has been
bilities, skills, and practices. DevOps moving to large-scale DevOps might practicing DevOps for eight years,
potentially amplifies the ­b oundary impact KS practices. being one of the very early adopt-
distortion when applied at a large In sum, the following KS chal- ers of the approach. We followed
scale. In this article, we focus on inno- lenges in DevOps have been identi- the case-study method of combining
vative practices for knowledge sharing fied in the literature: interviews, observations, and docu-
(KS), which is one of the four pillars, mentation. To identify the KS prac-
and argue that these approaches are 1. more intense cross-functional tices implemented by project teams,
more likely to emerge in large-scale collaboration between Dev we made direct observations during
DevOps. A KS practice occurs when and Ops10,11 field visits, attended 20 meetings plus
an individual transfers what he or she 2. multiple environment incompat- an eight-hour DevOps coaching day,
knows to another person. ibilities leading to specialized and conducted semi-structured in-
When moving to large-scale agile, teams12,13 terviews with 106 employees. The
the KS challenges and related suc- 3. the ability to self-organize14 teams included multiple functions as-
cess criteria have been identified,4,5 4. a loss of global vision of a proj- sociated with software development,
which raises the issue of ensuring ect and application knowledge such as developers, project managers,
and improving learning and KS prac- due to automation7 release engineers, user-experience
tices. Ghobadi et al. identified barri- 5. the confinement of KS to hierar- designers, the product owner (PO),
ers6 and risks to effective KS in agile chical organizational structure and architects.
teams. Risk perception is even higher and red tape7 For each project, three of the four
in DevOps than in agile.7 Conse- 6. limited sharing when parts of authors worked on a deep analysis.
quently, KS challenges are larger and the development and operations First, we collected information re-
more complex since more points of are outsourced.15 garding the agile methods and actual
view come into consideration when practices supporting the KS inside
using large-scale DevOps. Managing Moving from DevOps to large- and outside of the teams and related
knowledge dependencies becomes scale DevOps amplifies challenges to the use of space (that is, a com-
critical in a large-scale DevOps con- including coordination and collabora- mon work area), agile methodologies
text.8 Indeed, some individuals could tion improvement, dependency man- (for instance, Scrum-of-Scrum meet-
form a bottleneck to the knowledge agement, knowledge development, ings and the daily stand-up), and

M AY/J U N E 2 0 2 0 | I E E E S O F T WA R E 31
Authorized licensed use limited to: UNIVERSIDAD VERACRUZANA. Downloaded on February 19,2023 at 01:07:05 UTC from IEEE Xplore. Restrictions apply.
FOCUS: DevOps: LESSONS LEARNED

collaborative tools enabling the KS teams distributed across four geo- of the functions were performed by
(such as Slack and Atlassian JIRA). graphical sites and two remote coun- the same individuals. To overcome
Second, we compared our findings tries, which were distant at many this difficulty, one of the large proj-
with a baseline reference (the 12th levels, according to Ghemawat’s cul- ect teams at an advanced level of
“State of Agile Report”) to identify tural, administrative, geographic, DevOps maturity proposed a new
potential innovative practices. We and economic model.17 Development practice that involved a cross-func-
studied specific adaptations of agile activities were essentially based in tional DRR and, therefore, associ-
practices and the development of KS the same country, while the opera- ated responsibilities. Job rotation in
procedures that had emerged to suit tions were mostly in another. Proj- the software-development context
the contingencies of the development ect members chiefly used Scrum and was identified in a previous study18
environment. Third, we investigated Scrumban and scaled up with LeSS- as a means of addressing organiza-
the literature, where we found prac- adopting practices such as Scrum tional concerns and stimulating in-
tices that the company used and that of Scrum. They implemented an in- novation. The instantiation of the
were not present in the baseline ref- ternational, cross-functional daily role rotation in the cross-functional
erence report. DevOps meeting. DRR practice in our case enabled
Identifying innovative KS prac- large-scale learning and KS since
tices thus meant identifying those Innovative Practices all team members were able to per-
that were the most advanced within for KS form several roles and become more
the 18 projects we investigated. They We propose t h re e levels of K S knowledgeable.
were likely to be new to the firm, but practices based on three orders of DRR is usually limited to a spe-
we do not claim that they are new DevOps maturity (agile, continuous cific area, such as the development
to the world. However, they help integration, and continuous deliv- or operations team. A developer can
address challenges when moving ery) during the transition from agile become a Scrum master for a week
toward large-scale DevOps. Six of to large-scale DevOps (see Table 1). or sprint, then go back to his or her
the projects practiced DevOps, and At the first DevOps maturity level, position. In our case, the DRR was
innovative KS methods were found agile teams do not cross between cross functional, and, for instance,
in only two of the projects that functional silos, and DevOps is non- production engineers (operations)
used large-scale DevOps. Project X existent. At the second level, teams performed code reviews. All mem-
started eight years ago, operating in practice continuous integration; bers from both sides, development
DevOps mode from the beginning, hence, across silos. At the third level, and operations, were able to meet
and Project Y matured to operate in teams practice continuous delivery and work together alternatively.
DevOps mode in 2016. Both were and/or deployment. We identify four Cross-functional DRR ensured that
advanced in DevOps and practiced practices that emerged through tai- the “heavy lifting” workload did not
continuous delivery and deployment. loring the method and address the repeatedly fall on the same individu-
Project X involved 85 people or- needs of DevOps in a large-scale de- als. It also fulfilled the succession-
ganized in seven teams that were velopment context: cross-functional planning role: Team compositions
colo c ate d exc ept for a n op era- dynamic role rotation (DRR), tech- became more flexible since individu-
tions group a few miles away. The nical Thursdays (T Ts), heads-up als could be swapped out without
teams were not necessarily the same grooming and planning (H UGP), causing perturbation to the extent
size and did not have to be com- a nd t he ci rcle. We d is c uss each that the groups could not function.
posed the same way. Project mem- in turn.
bers worked in DevOps and mainly TTs
used a hybrid agile method derived Cross-Functional DRR Another practice that emerged in the
from Scrum, Kanban, and XP. They A significant involvement of certain context of technical KS was the TTs.
scaled up their agile method, first functions (project Manager, PO, Those “tech-talk” events took place
with large-scale Scrum (LeSS). One and Scrum master) was highlighted for half a day every two weeks. The
year later, they moved to the scaled during daily stand-up meetings. In goal was to disseminate technical
agile framework (SAFe). Project Y some projects, this involvement was knowledge, particularly in relation
involved 65 people divided into five even more pronounced as several to technologies, tools, skills, and

32 I E E E S O F T WA R E | W W W. C O M P U T E R . O R G / S O F T W A R E | @ I E E E S O F T WA R E

Authorized licensed use limited to: UNIVERSIDAD VERACRUZANA. Downloaded on February 19,2023 at 01:07:05 UTC from IEEE Xplore. Restrictions apply.
Table 1. Knowledge-sharing practices during the transition to large-scale DevOps
(after Hemon et al.15).
Maturity level Characteristic of KS practice Path to large-scale DevOps

Level 1: Agile There is a more frequent KS due to the iterative process and Large-scale agile methods are used (for example, SAFe and
more frequent releases. LeSS).
There is better communication and sharing between the DevOps is not achieved because silos still exist.
customer and developers.
The KS is limited to specific silos (mostly development and, to
There is limited sharing among the team and little common a lower level, business).
culture with the PO.

Level 2: Continuous Through the performance of various tests (unit and Some large-scale frameworks are used and meet their first
integration nonregression), the KS is synchronized with the code limits (such as the rigidity of SAFe) along the entire pipeline
development.
The alignment of the operations function with the
Task automation is achieved as far as possible, with development function is achieved.
knowledge transfer.
Some DevOps pillars, such as sharing, are partially reached.
There is partial KS. Developers need quick feedback.
The KS is applied across silos between development and
A partial common vocabulary and culture boosts the KS. operations.
There are some common tools fostering the KS.
There are different metrics and measurement systems
limiting the KS.

Level 3: Continuous Integration tests with other components, end-to-end Large-scale frameworks are particularly challenged on the KS
delivery tests, performance tests, and user-acceptance tests are along the entire pipeline.
codesigned, performed, and, preferably, automated by
There is a higher degree of alignment, sharing, and
operations in conjunction with the development function.
automation.
There is learning and extensive KS.
There are full DevOps practices based on the four pillars, with
There is a shared backlog. a high level of KS across teams and beyond.
There is a shared work system.
There are shared metrics and measurement tools.
Innovative KS practices appear.

other topics that would typically be The difficulty with the tech talks However, POs, architects, designers,
championed and led by an individual concerns the fact that the KS could developers, testers, and operations
team member. The TTs could take be limited to a group, person, or engineers were invited. Participants
the form of workshops on new top- community of practice (CoP). There- played for teams other than their
ics (for example, containerization, fore, inter-team KS at a large scale usual one. Teams had to be mixed
automated configuration manage- could be an issue for agile software- and represent as many trades as pos-
ment, and infrastructure as code), development organizations.19 The sible at the organization level. While
related tooling (Puppet and Docker), originality of the TT events is that the hackathons were frequented
fresh initiatives (A/B experimenta- they go beyond CoPs and the risk mostly by developers, the game fos-
tion), challenges (hackathons), and of developing knowledge on a tribal tered large-scale KS across the de-
games (Code Wars). A playful spirit basis. The TTs are fully cross func- velopment and operations areas and
and gamification were often strong tional both vertically and horizon- significantly helped the realization of
components, which had the extra tally across silos. One could have projects; for instance, the implemen-
benefit of boosting morale and the supposed that the Code War game tation of Docker containers for a de-
team spirit. was only intended for developers. ployment system that was proposed

M AY/J U N E 2 0 2 0 | I E E E S O F T WA R E 33
Authorized licensed use limited to: UNIVERSIDAD VERACRUZANA. Downloaded on February 19,2023 at 01:07:05 UTC from IEEE Xplore. Restrictions apply.
FOCUS: DevOps: LESSONS LEARNED

by operations and accepted by devel- than a discussion about user-story fulfilled the role of the sprint retro-
opment outside of the project teams. estimation. H UGP strongly pro- spective at a large scale by stimulat-
motes KS, pushing team members ing and animating reflection on the
HUGP to delve into the code, learn, test daily functioning of the teams and
Backlog grooming, or backlog refine- something new, enrich each other, organization, and it had the added
ment, can be understood in differ- and even consult operations engi- advantage of consisting of multiple
ent ways. In the projects we studied, neers on specific topics (such as the groups. This reflection was particu-
backlog grooming consisted of short- production environment and deploy- larly useful for implementing the
term user-story costing. However, in ment package). Before the discus- other innovative practices mentioned
a large project where the team used sion, there is a transfer of knowledge here. New practices do not always
DevOps, one PO initiated a prac- between members that helps them get everybody’s support; therefore,
tice to support backlog grooming to surface their arguments for bet- their implementation must be the
and sprint planning, which is labeled ter estimates. The benefits of such subject of open discussion.
HUGP. This practice is similar to KS practices are multiple at the When the shared knowledge ap-
rolling look-ahead planning (RLP) in DevOps-team level, then across the peared to be valuable, such as during
the sense that there is a high-level re- project: time-saving when estimating Code Wars, it was adopted and dis-
lease plan, and sprints are gradually user stories, clearer and more accu- seminated across teams. In that way,
refined as they start. In the HUGP rate estimations, and engendering a team members limited the problem
ceremony, the PO no longer performs greater sense of development-team of missing valuable knowledge or
any immediate user-story estima- empowerment. not capitalizing on it. The circle en-
tion with the team. Instead, the PO abled such discussion to take place
provides the team with broad infor- The Circle and ensured that innovative prac-
mation on the subject. The idea is to One issue with sprint retrospectives is tices could evolve, much in the spirit
give the team a visualization of the that, if they are done at all, they are of the quote by DeMarco. The prac-
work to come. The meeting is short, often combined with the sprint-review tice typically responded to the KS
somewhere between 10 and 30 min. ceremony. Also, they are confined to a challenge when large-scale DevOps
Subsequently, the sprint-planning single team, thereby limiting the KS19 was applied (see Table 1) and went
meeting takes place and lasts two and minimizing its opportunity to play beyond DevOps boundaries by also
hours during which user stories are a role in achieving continuous improve- integrating the business domain.
presented, estimated, and broken ment. Consequently, agile and DevOps
down into tasks. However, the effect are not fostered at a large-scale. Resolution of Major KS Challenges
of the previously held HUGP practice The team met difficulties with Different innovative practices have
is very noticeable. The PO describes retrospectives, primarily because the met the KS challenges in the context
it well: “The team is more confident forums were quite complex to orga- of large-scale DevOps. In Table 2, we
and self-assured when it comes to nize with 85 people. The members summarize the main difficulties and
story-point estimation. There is less decided to set up joint-reflection cer- impacts that each of them generates.
error or difference. The team mem- emonies, which were known as the
bers have had time to go through the circle. The practice was based on a Large-Scale DevOps and
code and figure out if they had any discussion among a group of people Innovative KS Practices
doubts. They discuss it among them- that represented all of the jobs and While the Scrum method was the
selves, around a foosball game.” roles across multiple project teams, backdrop for the projects we sur-
We consider HUGP as an ex- including managers, POs, architects, veyed, reality is much more complex
ten sion of R L P. B ot h adopt a developers, testers, and operations in terms of overlapping and comple-
just-in-time approach to up-front engineers. Participation was volun- mentary practices.20 Thus, we were
planning; both foster discussion tary. The circle was a forum for the able to identify the development of
within teams for the identification real exchange of knowledge across innovative KS practices only in the
and estimation of product backlog team functions. It was composed of most advanced projects (see Table 1).
components. However, we suggest approximately 20 people who met This raises the question of how large-
that HUGP differs because it is more twice a month. The circle practice scale DevOps enables innovation.

34 I E E E S O F T WA R E | W W W. C O M P U T E R . O R G / S O F T W A R E | @ I E E E S O F T WA R E

Authorized licensed use limited to: UNIVERSIDAD VERACRUZANA. Downloaded on February 19,2023 at 01:07:05 UTC from IEEE Xplore. Restrictions apply.
Table 2. The major challenges and impacts of innovative practices.
Innovative Major challenges
Impact
practice addressed

Cross- • Intense cross-functional The breaking of knowledge dependencies is enabled15 because information is not held by any
functional DRR collaborations individual. However, this obliges team members to communicate to carry out their roles effectively.
• Team self-organization When team members rotate, they can take on responsibilities, develop skills, and acquire
challenges knowledge. This fosters the team’s autonomy.

TTs • More intense cross- Team participants (representing almost every area, such as business, development, quality, and so
functional collaboration forth) can discover and learn new approaches without any consideration of belonging to a specific
functional silo or the position held. This is how operations teams bypassed the organizational
culture related to training, and operations made development teams discover Rancher, an open-
source tool to deploy and manage containers in production environments.
• Complex environments Teams are helped to avoid over-specializing through fostering the KS and increasing the number of
and specialized teams individuals who understand the multiple environmental complexities and incompatibilities.

HUGP • Confinement of the KS The KS is facilitated at scale, avoiding additional red tape and reducing the burden of agile
due to the hierarchical and processes and organizational rules and procedures.
organizational structure
By practicing HUGP, the manager gives the team members the possibility to obtain central
and red tape.
information that they would not have had, while respecting the processes of the agile method and
organization. Team members can, therefore, use information in advance, which enables them to
enrich and transform the information into knowledge. This new knowledge is shared, helping team
members make more realistic and accurate decisions during their mission.

The circle • Capability to self-organize The circulation of knowledge is facilitated because the circle is fully cross-functional, with
representatives from all of the jobs linked to the project and, at the same time, because it consists
of only 20 individuals representing larger teams (85 people).
• Multiple environment This fosters the KS at a large scale, decision making, and, consequently, teams’ autonomy.
incompatibilities leading
teams’ specialization.

Our hypothesis is that these innova- of deliverables), and they used basic the business, development, and opera-
tion practices are the result of further practices. However, the KS was less tions functions. Such KS methods suit
continuous improvement facilitated developed than it was for the other projects operating at a large-scale and
by stronger levels of self-organization two projects. This may be because the in DevOps because those joint condi-
in DevOps teams acting on a broader four projects had less time to evolve tions multiply the coordination issues
mandate. As mentioned, these KS the practice and because three of and risks related to a lack of compe-
practices were observed in large-scale them were smaller initiatives. tencies.15 Moreover, we believe that
projects that commenced eight years Interestingly, in this large company, project autonomy, or self-organization,
ago, one of them operating in DevOps the innovative KS practices were not and continuous improvement during a
mode from the beginning, and the shared across all projects, even at the long period of time are critical influ-
other maturing to eventually oper- same software-process maturity level. encing factors. Thus, in the teams that
ate in DevOps mode. The other four The creative KS practices addressed developed KS practices:
DevOps projects were clearly shar- specific challenges that projects faced
ing knowledge (for instance, com- and helped the organization to par- • Projects were more advanced in
mon project management, processes, tially or fully resolve initial problems. agile and tended toward large-
and tools, such as shared continuous Large-scale DevOps may require more scale DevOps.
integration and automated deploy- KS practices since additional con- • The larger the project teams were
ment tools and the co-construction straints have to be integrated between the more they self-organized and

M AY/J U N E 2 0 2 0 | I E E E S O F T WA R E 35
Authorized licensed use limited to: UNIVERSIDAD VERACRUZANA. Downloaded on February 19,2023 at 01:07:05 UTC from IEEE Xplore. Restrictions apply.
FOCUS: DEVOPS: LESSONS LEARNED
ABOUT THE AUTHORS

AYMERIC HEMON is an assistant profes- BARBARA LYONNET is an associate


sor in management information systems professor of management sciences at the
(ISs) at the ESSCA School of Management, University of Nantes, France, where she
Angers, France, and a researcher at the is also a researcher in the Laboratoire
Laboratoire d’Economie et de Management d’Economie et de Management de Nantes
de Nantes Atlantique, University of Nantes, Atlantique. Lyonnet received a Ph.D. in
France. He is also a senior lecturer in industrial engineering from the University
software engineering at the Mines Telecom of Savoie Mont-Blanc, France, in 2010. She
Institute, Paris, France. His research has published in the field of information
focuses on IS-enabled organizational systems and quality and supply-chain man-
transformation, particularly agile methods agement, including several books, such as
and DevOps. Hemon received a Ph.D. in Lean Management. Contact her at barbara.
management information systems from lyonnet@univ-nantes.fr.
the University of Nantes, France, in 2018.
Contact him at aymeric.hemon-hildgen@
essca.fr.

BRIAN FITZGERALD is the director of FRANTZ ROWE is a professor at the Uni-


Lero, a software research center at the versity of Nantes, France, where he is also
University of Limerick, Ireland. He also a researcher at the Laboratoire d’Economie
holds an endowed professorship, the et de Management de Nantes Atlantique.
Krehbiel Chair in Innovation in Business and He is a professor at the SKEMA Business
Technology, at the University of Limerick. School, Sophia Antipolis, France; former
Fitzgerald received a Ph.D. in computer chief information officer of the University
science from the University of London in of Nantes; and professor at Telecom Paris.
1997. His publications include 15 books Rowe received a Ph.D. in the management
and more than 150 peer-reviewed articles of information systems from the University
in international journals and conferences of Paris X, Nanterre, France, in 1991. He
related to the information-systems and is an Association for Information Systems
software-engineering fields. Contact him fellow and editor emeritus of the European
at brian.fitzgerald@ul.ie. Journal of Information Systems. He directs
the research program on DevOps at the
University of Nantes. Contact him at frantz.
rowe@univ-nantes.fr.

used their relative autonomy to


continuously improve throughout
a long period.
• Project teams were more likely
to develop innovative KS
C learly, the emergence and
implementation of these in-
novative KS practices is
an outcome of multiple idiosyn-
cratic conditions, and we would
in our ongoing research at other
firms tend to support a large-scale
DevOps effect and continuous-
improvement result related to self-
organization in the generation of
practices among their mem- need data beyond the 18 projects in novative K S prac tices. Using
bers and across functional that we studied to reach any firm DevOps at a large-scale amplifies
boundaries. conclusion. However, observations the KS challenge as well as related

36 I E E E S O F T WA R E | W W W. C O M P U T E R . O R G / S O F T W A R E | @ I E E E S O F T WA R E

Authorized licensed use limited to: UNIVERSIDAD VERACRUZANA. Downloaded on February 19,2023 at 01:07:05 UTC from IEEE Xplore. Restrictions apply.
solutions through the development 7. A. Hemon, L. Monnier-Senicourt, and teams—How incumbent companies
of innovative practices. Such effects F. Rowe, “Job satisfaction factors and achieve competitive advantages,”
are not fixed since the implementa- risks perception: An embedded case in Proc. 51st Hawaii Int. Conf.
tion of these practices is not rigid, study of DevOps and agile teams,” System Sciences (HICSS), 2018,
and they are subject to continu- in Proc. 39th Int. Conf. Information pp. 4931–4940. doi: 10.24251/
ous adjustment to avoid deteriora- Systems (ICIS), San Francisco, 2018. HICSS.2018.617.
tion and rigidity or, indeed, being 8. V. Stray, N. B. Moe, and A. Aasheim, 15. A. Hemon, B. Lyonnet, F. Rowe,
poured into concrete. “Dependency management in large- and B. Fitzgerald, “Conceptualizing
scale agile: A case study of DevOps the transition from agile to DevOps:
References teams,” in Proc. 52nd Hawaii Int. A maturity model for a smarter IS
1. P. Debois, “DevOps: A software Conf. System Sciences (HICSS), function,” in Proc. Int. Working
revolution in the making,” J. Inform. Grand Wailea, HA, 2019. doi: Conf. Transfer and Diffusion IT
Technol. Manage., vol. 24, no. 8, 10.24251/HICSS.2019.840. (TDIT), 2018, pp. 209–223. doi:
pp. 3–39, Aug. 2011. 9. P. A. Nielsen, T. J. Winkler, and J. 10.1007/978-3-030-04315-5_15.
2. M. Virmani, “Understanding Nørbjerg, “Closing the IT develop- 16. Y. Zhao, A. Serebrenik, Y. Zhou,
DevOps & bridging the gap from ment-operations gap: The DevOps V. Filkov, and B. Vasilescu, “The
continuous integration to continu- knowledge sharing framework,” impact of continuous integration
ous delivery,” in Proc. 5th Int. Conf. in Proc. 16th Int. Conf. Perspec- on other software development
Innovative Computing Technology tives Business Informatics Research practices: A large-scale empiri-
(INTECH), 2015, pp. 78–82. doi: (BIR), Copenhagen, Denmark, 2017, cal study,” in Proc. 2017 32nd
10.1109/INTECH.2015.7173368. vol. 1898, chap. 5, pp. 1–15. IEEE/ACM Int. Conf. Auto-
3. T. Dingsøyr, T. E. Fægri, and J. 10. W. Gottesheim, “Challenges, ben- mated Software Engineering
Itkonen, “What is large in large- efits and best practices of per- (ASE), pp. 60 –71. doi: 10.1109/
scale? A taxonomy of scale for formance focused DevOps,” in ASE.2017.8115619.
agile software development,” in Proc. 4th Int. Workshop Large- 17. P. Ghemawat, “Distance still mat-
Proc. Int. Conf. Product-Focused Scale Testing, 2015, pp. 3–3. doi: ters,” Harv. Bus. Rev., vol. 79, no. 8,
Software Process Improvement 10.1145/2693182.2693187. pp. 137–147, Sept. 2001.
(PROFES), 2014, pp. 273–276. doi: 11. A. Wiedemann, “A new form of col- 18. T. E. Fægri, T. Dybå, and T. Ding-
10.1007/978-3-319-13835-0_20. laboration in IT teams—Exploring søyr, “Introducing knowledge
4. T. Dingsøyr and N. B. Moe, “Re- the DevOps phenomenon,” in Proc. redundancy practice in software
search challenges in large-scale 21st PACIS Asia Pacific Conference development: Experiences with job
agile software development,” ACM Information Systems, 2017. pp. 1–12. rotation in support work,” Inf. Softw.
SIGSOFT Softw. Eng. Notes, vol. [Online]. Available: http://aisel.aisnet Technol., vol. 52, no. 10, pp. 1118–
38, no. 5, pp. 38–39, Sept. 2013. doi: .org/pacis2017/82 1132, Oct. 2010. doi: 10.1016/j.
10.1145/2507288.2507322. 12. L. E. Lwakatare et al., “Towards infsof.2010.06.002.
5. D. Smite, N. B. Moe, G. Levinta, and DevOps in the embedded systems 19. V. Santos, A. Goldman, and C.
M. Floryan, “Spotify guilds: How domain: Why is it so hard?” in Proc. R. De Souza, “Fostering effective
to succeed with knowledge sharing 2016 49th Hawaii Int. Conf. System inter-team knowledge sharing in
in large-scale agile organizations,” Sciences (HICSS), pp. 5437–5446. agile software development,” Em-
IEEE Softw., vol. 36, no. 2, pp. 51– doi: 10.1109/HICSS.2016.671. pir. Softw. Eng., vol. 20, no. 4, pp.
57, Mar.–Apr. 2019. doi: 10.1109/ 13. J. Smeds, K. Nybom, and I. Porres, 1006–1051, Aug. 2015. doi: 10.1007/
MS.2018.2886178. “DevOps: A definition and perceived s10664-014-9307-y.
6. S. Ghobadi and L. Mathiassen, adoption impediments,” in Proc. 20. B. Fitzgerald, G. Hartnett, and K.
“Perceived barriers to effective Int. Conf. Agile Software Devel- Conboy, “Customising agile methods
knowledge sharing in agile software opment, 2015, pp. 166–177. doi: to software practices at Intel Shan-
teams,” Inform. Syst. J., vol. 26, 10.1007/978-3-319-18612-2_14. non,” Eur. J. Inf. Syst., vol. 15, no.
no. 2, pp. 95–125, Dec. 2014. 14. A. Wiedemann, “IT gover- 2, pp. 200–213, Jan. 2006. doi:
doi: 10.1111/isj.12053. nance mechanisms for DevOps 10.1057/palgrave.ejis.3000605.

M AY/J U N E 2 0 2 0 | I E E E S O F T WA R E 37
Authorized licensed use limited to: UNIVERSIDAD VERACRUZANA. Downloaded on February 19,2023 at 01:07:05 UTC from IEEE Xplore. Restrictions apply.

You might also like