| Course | BU 455 Big Data Management |
|---|---|
| Unit | Unit 8 |
| Paper type | Architecture recommendation |
| Length | About 920 words, 6 pages |
| Format | APA 7 student paper |
| School | Herzing University |
| Program | BS in Business Analytics |
| Updated | October 2026 |
Free sample paper for BU 455 Unit 8
Keep the Database, Add a Replica, Files and a Small Alert Path: An Architecture Recommendation for an Invented City Bike Service, and the Evidence That Would Prove It Wrong
Student Name
Business, Herzing University
BU 455: Big Data Management
Instructor Name
April 22, 2026
Keep the Database, Add a Replica, Files and a Small Alert Path: An Architecture Recommendation for an Invented City Bike Service, and the Evidence That Would Prove It Wrong
The Requirement Being Answered
Pedalworks Mobility, the composite bike-share operator this course has followed, asked whether it needs a big data platform. The question behind that request, drawn from the term's work, is narrower and more useful. The company needs an architecture that can alert crews within minutes to theft, faults, dangerous batteries and full or empty stations; let analysts query years of history without slowing those alerts; reconcile telemetry arriving in several formats; delete data on the schedule the governance plan sets; and cope if bikes start reporting every five seconds, which would multiply location data about sixfold. This recommendation answers that requirement for 2026 through 2028.
The recommendation is Option A from the cost and capability analysis: keep the managed PostgreSQL database for operations, add a read replica for analysts, move telemetry older than 90 days to monthly columnar files queried from a single machine, and put the four real-time rules on a rented queue feeding one lightweight alert program. Almost every component is either already in place or a standard managed service, and that is deliberate.
The Recommended Design Against the Rejected One
The strongest alternative was Option C, the distributed platform the vendor originally proposed. The table sets the two side by side against each part of the requirement.
| Requirement | Recommended: extended conventional | Rejected: distributed platform |
|---|---|---|
| Alerts within minutes | Queue and a small service; 95% of alerts in about 4 seconds in testing | Streaming engine; similar speed with far more components |
| History without slowing operations | Replica and summary tables; battery question in 3 seconds on the summary, 9 minutes as a full scan | Cluster; full scan about 3 minutes, but only after tuning for skew |
| Mixed formats | One translation view, weekly layout check | Same translation needed, written for the cluster |
| Retention and deletion | Scheduled jobs in the database and storage expiry rules | Jobs across more storage layers |
| Five-second telemetry | Absorbed with a larger replica and columnar archive | Absorbed with room to spare |
| Three-year cost | About $193,800 | About $662,600 |
| New skills needed | One, learnable in about two weeks | Two, requiring a new hire |
The distributed platform would do everything the recommended design does, and it would handle growth far beyond the company's current path. It is rejected because Pedalworks does not have that growth, and because its cost falls mostly on people the company does not have. The distributed processing exercise showed a single workstation finishing the key analytical job faster than the cluster at today's volume. McSherry et al. (2015) found that many distributed systems need large clusters merely to match a single well-programmed machine, and Pedalworks' test is a small instance of the same finding.
What the Change Involves
The work takes about six weeks of the data engineer's time: two weeks to set up the replica, partitions and summary tables; two weeks for the monthly export to columnar files and the analysts' training; and two weeks for the queue, the alert service and its failure handling. Running costs rise from about $10,000 to about $21,300 a year, mostly for the replica and the queue. Over three years, including staff time, the total is about $193,800, against about $273,400 for a managed warehouse and $662,600 for the distributed platform. One analyst will be trained on the nightly jobs so that the design does not depend on a single person.
What the Vendor's Proposal Got Right
The vendor's proposal deserves credit on two points, and the recommendation adopts both in smaller form. It was right that operations needed faster alerts than the 15-minute reports provided; the queue and alert service meet that need. And it was right that analysts needed a better way to work with history than querying the operational database; the replica and columnar files meet it. Where the proposal went wrong was in the scale of its answer, a platform designed for data volumes many times larger than Pedalworks has, built from components the company could not run. Recording what the rejected option got right keeps the recommendation from reading as a dismissal, and it makes clear which needs would still be served if the company later outgrows the recommended design.
How the Change Will Be Rolled Out
The work proceeds in three steps, each reversible. The replica and summary tables go first, because they protect operations from analysts' queries immediately and change nothing else. The columnar archive follows once the analysts have been trained, with the database keeping its full telemetry for one extra month so that any error in the export can be caught before old rows are removed. The alert service goes last and runs in parallel with the existing 15-minute reports for two weeks, so crews can compare the two and trust the new alerts before the old reports are retired. Each step has an owner and a check that must pass before the next begins.
Evidence That Would Make This Recommendation Wrong
A recommendation should say what would overturn it. Five observations would, and each is measurable from systems the company already runs. First, if telemetry grows past about ten times its 2025 volume, for example through the five-second firmware combined with a doubled fleet, full scans on one machine would become too slow, and the distributed option should be revisited. Second, if full-history scans on the replica regularly exceed 30 minutes, analysts' work would suffer. Third, if 95 percent of alerts take longer than two minutes, the alert path would need redesign. Fourth, if the city's contract changes to require a live, second-by-second data feed, the streaming side would need to grow. And fifth, if the data engineer leaves and the trained analyst cannot keep the jobs running, the operational simplicity argument would weaken and a managed warehouse might suit better.
Each quarter the chief executive receives these five figures on a single page. Stonebraker and Cetintemel (2005) argued that systems should be matched to their workloads, and Kleppmann (2017) that simplicity is worth protecting until a workload truly demands otherwise. Recommending little that is new is an acceptable answer when the evidence supports it, and this evidence does. Pedalworks and every figure in this recommendation are invented for the course.
References
Kleppmann, M. (2017). Designing data-intensive applications: The big ideas behind reliable, scalable, and maintainable systems. O'Reilly Media.
McSherry, F., Isard, M., & Murray, D. G. (2015). Scalability! But at what COST? In 15th Workshop on Hot Topics in Operating Systems (HotOS XV). USENIX Association.
Stonebraker, M., & Cetintemel, U. (2005). "One size fits all": An idea whose time has come and gone. In 21st International Conference on Data Engineering (pp. 2-11). IEEE. https://doi.org/10.1109/ICDE.2005.1
BU 455 Unit 8 instructions, in plain terms
The final BU 455 unit asks students to recommend a data architecture for the case. Typical requirements include stating the requirement the architecture must meet, describing the recommended design, comparing it directly with the strongest rejected alternative, pricing the change and naming the evidence that would make the recommendation wrong. Instructors routinely confirm that a recommendation to make no major architectural change can earn full credit if the earlier units back it. The design section should be detailed enough to act on but need not specify every configuration. Writing the recommendation for a real employer is discouraged. A comparison table and a concise APA 7 recommendation, drawing on the earlier units, are typical.
How this BU 455 Unit 8 example is built
The recommendation opens by restating the requirement behind the request for a platform and naming the recommended design. A table compares it with the rejected distributed platform on seven points, from alert speed to skills, followed by the reasons for rejection and a cited finding about single-machine baselines. A section credits two things the vendor's proposal got right. A section describes the work, its cost and the three-year totals. A section lays out a three-step rollout, each reversible, with checks between steps. The final section names five measurable observations that would overturn the recommendation and commits to reporting them quarterly.
Where the points sit in the BU 455 Unit 8 rubric
Architecture recommendation rubrics at Herzing generally reward a clear requirement, a specific recommended design, a direct comparison with the rejected option, priced change and falsifiable conditions, with writing and APA. The requirement criterion earns full credit when needs are stated before the design. The design scores highest when components and their roles are clear. Comparison rewards setting the recommendation against the strongest alternative on the same terms. Pricing is judged on build effort and ongoing costs. Falsifiable conditions earn points when they are measurable and tied to a review. Instructors also credit recommendations that acknowledge what the rejected option got right, since that shows balanced judgment.
BU 455 Unit 8 help before you submit
Architecture recommendations often lose points by describing the chosen design without comparing it to anything. Another common problem is recommending a platform because it is popular rather than because the requirements demand it. Students frequently omit costs or give infrastructure prices without staff time. Some present a recommendation as permanent, with no conditions for revisiting it. Others dismiss the rejected option without crediting its strengths. Many skip the rollout, leaving no path from current to recommended state. Steps that can each be undone, with a check before the next one starts, show a reader that the recommendation was planned as work and not only argued as an idea. Finally, writing the recommendation for your own employer can expose internal plans; the bike-share case demonstrates identical reasoning and exposes nobody.
Get a BU 455 Unit 8 example written to your instructions
Collect the Unit 8 instructions, the BU 455 rubric and the earlier work this recommendation is meant to build on, then send the lot across. The custom example puts the recommendation near the front, keeps the rejected option beside it, phases the change, and states the conditions that would overturn the call. First custom sample free, returned in 24 to 48 hours.
More BU 455 papers and business analytics samples
- BU 455 Unit 1: Scale assessment
- BU 455 Unit 2: Conventional alternative analysis
- BU 455 Unit 3: Storage and format decisions
- BU 455 Unit 4: Batch and streaming design
- BU 455 Unit 5: Distributed processing exercise
- BU 455 Unit 6: Governance and retention plan
- BU 455 Unit 7: Cost and capability analysis
- BU 451 Unit 8: Reproducible analysis submission
- BU 453 Unit 8: Database design project
- BU 454 Unit 8: Visual report with rationale
- BU 351 Unit 8: End-to-end analytics case
BU 455 Unit 8 questions, answered
Is recommending no architectural change an acceptable answer?
Yes, when the document earns it. Carry the requirement through, demonstrate that the design already in place satisfies it, and name the growth or new obligation that would force a rebuild. The weak form of this answer is the one that never took a distributed option seriously, because a marker cannot tell a considered verdict from a writer who ran short of time.
How much detail does the design section need?
Enough for a reader to judge whether it meets the requirement, and no more. Components, what each is responsible for, how records pass between them and who operates them is usually the level these sections want. Configuration detail and version numbers date quickly and rarely earn anything, while an unanswered question about who runs a component overnight will cost you.
Could I write the recommendation for my own employer?
Not as the submitted document. A recommendation of this kind names the systems the organization runs today, what they cost, where they fall short and what replacing them would take, which is a package no company wants in circulation. Write it against the case. Nothing stops you carrying the reasoning into a proposal at work later, on your employer's terms and in their own format.
Where can I find a free BU 455 Unit 8 sample paper?
The complete BU 455 Unit 8 architecture recommendation can be read here for free, with the comparison table, the rollout and margin notes. If your case differs, send the assignment and we write a custom recommendation at no cost, typically within 48 hours.
Is recommending no architectural change acceptable in BU 455 Unit 8?
Yes, if the evidence supports it. The sample recommends keeping the existing database with modest additions and names five measurable conditions that would justify a larger change later.