| Course | BU 455 Big Data Management |
|---|---|
| Unit | Unit 3 |
| Paper type | Storage and format decisions |
| Length | About 1,000 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 3
Six Decisions, Each Against Something: Storage and Format Decisions for a Composite Bike-Share Operator's Telemetry, Photos and Station History
Student Name
Business, Herzing University
BU 455: Big Data Management
Instructor Name
February 11, 2026
Six Decisions, Each Against Something: Storage and Format Decisions for a Composite Bike-Share Operator's Telemetry, Photos and Station History
What Prompted These Decisions
The conventional alternative analysis showed that Pedalworks Mobility, the invented bike-share operator this course follows, can meet today's needs with its PostgreSQL database, a read replica and summary tables. It also identified the change that would strain that arrangement: the bike maker's proposed five-second telemetry, which would push five years of history toward 1.2 terabytes and make full scans take close to an hour. This document makes the storage and format decisions that prepare for that change. Each decision names the option it displaced, states the property of the workload that separated the two, and admits what the choice makes harder later when someone reads the data back.
Kleppmann (2017) describes storage choices as trade-offs between how data are written and how they are read; a format that makes one access pattern fast usually makes another slower. The decisions below are organized around the access patterns Pedalworks actually has.
Decision 1: Old Telemetry Moves to Columnar Files
Chosen: telemetry older than 90 days moves from database rows to columnar files in the company's cloud object storage, in the open Parquet format. Displaced: keeping all telemetry as rows in PostgreSQL. Workload property: analysis of old telemetry reads a few fields, bike, time and battery, across very many messages, while operations only ever need the last few days. A columnar format stores each field together, so a query reading three fields out of twenty reads roughly three-twentieths of the data, and similar values stored together compress well. Abadi et al. (2008) showed that column-oriented storage can outperform row-oriented storage by large margins on this kind of analytical query. In a test on one year of messages, the Parquet files took about 4 gigabytes against about 37 gigabytes of raw JSON, and the battery-degradation question ran in 40 seconds with a lightweight query engine on an analyst's laptop.
What it makes harder: individual messages cannot be updated in place, a single message is slower to look up than in the database, and analysts must learn a second query tool alongside the database. The analytics lead estimated two weeks for the team to become comfortable with it, a cost worth stating because it falls on people rather than on the budget.
Decision 2: Files Hold the Unified Layout, and Raw Originals Are Kept for a Year
Chosen: the Parquet files store telemetry in the single unified layout produced by the translation view, with battery always as a percentage. Displaced: storing each message exactly as the bike sent it and translating at read time. Workload property: every analysis needs the unified fields, and translating three layouts on every read repeats the same work and the same risk of error. What it makes harder: if the translation contains a mistake, the stored files contain it too. For that reason the raw JSON messages are also kept, compressed, for one year, costing about 8 gigabytes, so the files can be rebuilt if a translation error is found.
Decision 3: Monthly Partitions and Large Files
Chosen: files are organized by month and written in sizes of about 128 to 256 megabytes. Displaced: one file per bike per day, which a first draft of the export produced, creating more than a million small files a year. Workload property: queries filter by date range and read many bikes at once, and query engines slow down sharply when they must open many tiny files. What it makes harder: the current month is incomplete until it closes, so questions about the last few weeks go to the database, not the files.
Decisions 4 to 6: Photos, Operational Data and Station History
Decision 4, photos: damage photos are stored as image files in object storage, with a row in the database holding the bike, ticket, date and file location. Displaced: storing images inside the database. Property: images are large, read rarely and never searched by their content, so they gain nothing from the database but make its backups much bigger. Harder later: the image and its record live in two places, so deleting a photo under the retention plan must remove both.
Decision 5, operational data: trips, payments and the last 90 days of telemetry and station status stay as rows in PostgreSQL. Displaced: moving everything to the file store. Property: operations read and write single records constantly and need transactions, which row storage in a database handles well and file storage does not.
Decision 6, station history: station status older than 90 days is kept at five-minute resolution rather than every minute. Displaced: full-resolution history. Property: planning questions about old station use, such as how often a station was empty at 8 a.m. in March, are answered well at five-minute intervals, and the change cuts station history by 80 percent. Harder later: minute-level questions about old periods can no longer be answered. The operations lead agreed to this in writing, since it is a real loss.
What These Decisions Deliberately Do Not Do
None of the six decisions introduces a distributed processing platform, and none moves operational data out of the database. The file store is ordinary cloud object storage the company already pays for, and the query engine used on the files runs on a single machine. The Apache Parquet documentation describes the format as designed for efficient storage and retrieval in analytical workloads regardless of processing framework (Apache Software Foundation, n.d.), which is why it suits a small company that may never need a cluster. If the five-second firmware arrives, these decisions absorb it; if it does not, they still halve the database's growth. Each decision is recorded in a short decision log with its date and owner, so that when a later analyst finds minute-level station data missing for 2025, the reason and the person who agreed to it can be found in a minute. The bike-share company and every number here are invented for the course.
References
Abadi, D. J., Madden, S. R., & Hachem, N. (2008). Column-stores vs. row-stores: How different are they really? In Proceedings of the 2008 ACM SIGMOD International Conference on Management of Data (pp. 967-980). Association for Computing Machinery. https://doi.org/10.1145/1376616.1376712
Apache Software Foundation. (n.d.). Apache Parquet documentation. https://parquet.apache.org/docs/
Kleppmann, M. (2017). Designing data-intensive applications: The big ideas behind reliable, scalable, and maintainable systems. O'Reilly Media.
The Herzing BU 455 Unit 3 instructions
The third BU 455 unit asks students to decide how and where data should be stored. Requirements usually include several storage and format decisions, each justified by workload properties such as read patterns, write frequency and data type, an explicit comparison with the rejected option, and an honest note of the costs or limits each choice creates. Many instructors ask how many decisions to cover and whether to name specific formats; naming open formats such as Parquet is common and helpful. Describing an employer's storage arrangements is discouraged. The submission is typically a decision-by-decision APA 7 document, often with measured or estimated effects such as size and query time.
How the BU 455 Unit 3 example is put together
The paper opens with the trigger for these decisions, the firmware change identified in Unit 2, and the idea that storage choices trade write speed against read speed. Decision 1 moves old telemetry to columnar files, with measured size and query time and a cited study. Decision 2 stores the unified layout and keeps raw originals for a year. Decision 3 sets monthly partitions and file sizes after a draft produced a million files. A combined section covers photos, operational data and station history, each with its displaced option and cost. The final section explains what the decisions deliberately avoid and notes the decision log that records each choice.
BU 455 Unit 3 rubric: what full points look like
Storage decision rubrics at Herzing typically reward decisions tied to workload properties, explicit rejected alternatives, honest costs, evidence of effect and consistency with earlier findings, plus writing and APA. Workload ties earn full credit when each choice follows from how data are written and read. Rejected alternatives score highest when named and compared directly. Costs are judged on whether the student states what becomes harder, not only what improves. Evidence rewards measured or estimated effects such as storage size or query time. Consistency earns points when decisions respond to problems found in earlier units. Instructors also credit written agreement for decisions that lose information, such as downsampling history.
BU 455 Unit 3 help: mistakes that cost points
Storage decision papers commonly lose points by listing technologies without saying what each replaced or why. Another frequent problem is claiming a format is better without naming the workload it serves. Students often ignore the costs of a choice, such as columnar files being hard to update. Some produce many tiny files, a well-known performance problem. Others store images inside databases, inflating backups. Many downsample or delete history without anyone agreeing to the loss. Writers also forget to keep raw originals for a period after a format change, so a translation error found later can never be repaired from the source. Finally, describing how your own company stores its data can reveal internal architecture; a composite case shows the same reasoning safely.
Get a BU 455 Unit 3 example written to your instructions
Bring us the Unit 3 instructions, the BU 455 rubric your section posts, and the scenario the assignment builds on. The custom example restates the workload in storage terms, writes every choice against the option it displaced, settles the reading side first, and attributes each capability figure to a dated source. First custom sample free, back 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 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 455 Unit 8: Architecture recommendation
- BU 453 Unit 3: Normalization exercise
- BU 454 Unit 3: Comparison and ranking charts
- BU 351 Unit 3: Descriptive analysis
- BU 407 Unit 3: Forecast selection record
BU 455 Unit 3 questions, answered
How many decisions should the document actually cover?
Fewer than the field offers and more than one. Three or four decisions argued properly, each carrying its displaced alternative and its accepted cost, meet the criterion better than eight mentioned in passing. If a choice does not follow from something in the workload description, it probably belongs to a different case, and cutting it leaves room to argue the ones that do belong.
Am I expected to name particular formats?
Follow the instructions, since sections differ on how concrete they want the answer. Where a name is expected, tie it to the property of the workload that made it fit rather than to its popularity, and give a documentation date if you quote a capability. Where the instructions stay general, arguing at the level of layout families is safer and usually reads as the stronger reasoning.
Can I describe how my company stores its data?
Keep it out of the document entirely. The layouts, the partitioning and the volumes inside an employer's platform are its architecture, they are commonly covered by something you signed on your first day, and a graded file travels well beyond the building. Argue the supplied case. Knowing what really slows a read down will keep your accepted costs realistic, and that judgment is yours to bring.
Where can I find a free BU 455 Unit 3 sample paper?
The BU 455 Unit 3 storage and format decisions example is free to read in full here, with all six decisions, their costs and margin notes. If your workloads differ, share the assignment and a matching set of decisions is prepared free, usually within two days.
Should BU 455 Unit 3 name particular formats?
Naming open formats helps, as long as each is justified by the workload. The sample names Parquet for analytical history because its column layout reads only the needed fields and compresses well.