---

Audience

The dashboard serves three stakeholder groups:

---

Dataset

Backblaze hard drive telemetry data between January 1st and 31st, 2025.

Link To Dataset

---

KPIs

---

Dashboard Structure

The dashboard uses a multi-view layout with a top-level "Select View" (top left) dropdown that switches between different graphs. Five key metric cards are always visible at the bottom. This design serves all three audience groups: operations managers get a quick perspective on health summaries, procurement teams can explore vendor/model comparisons, and data scientists can explore reliability patterns.

---

Metric Cards → KPI's

---

Filtering

---

Data Cleaning & Pre-processing

Decoding Model Numbers and Drive Size

---

Mapping SMART codes

Critical Failure Predictors

Operational Metrics

---

Dataset reduction

A python script was used to remove the majority of the columns in the dataset. This reduced the size of the files by about 75%.

---

Design Choices

Color Selection

The dashboard's colors serve three purposes:

---

Chart & Layout Rationale

---

Reflection

Most of this project involved learning Tableau and dealing with its issues. I went through the DataCamp course and learned the basics, but getting comfortable building the dashboard took significant time. Once I understood how to structure the data sources and panels, the interactive filtering worked well. It's clear (besides the errors I experienced) why people use it for this kind of work.

The main issue was Tableau crashing frequently. Loading more than about 7 days of data would cause the application to lock up or crash entirely. I capped the dataset at 9 days to have a working version. Additionally, the project got corrupted at one point and I lost a week of work. I also encountered UI bugs like the Data Source tab becoming unclickable. It was incredibly frustrating to work through these stability problems.

If I were to do this again, I would build it as a web dashboard instead. I have much more experience with JavaScript, HTML, and CSS, and I could have built something comparable significantly faster. Tableau works for certain use cases, but in this instance, it felt unnecessary, especially given the instability issues (I would get a new error every single time I would open the project).

If I had more time I would want to try connecting to a database and spend a lot more time extracting information from the dataset.

---

Code

Pre-processing code

# Columns to keep
columns_to_keep = [
    'date',
    'serial_number',
    'model',
    'capacity_bytes',
    'failure',
    'datacenter',
    # SMART raw values only
    'smart_5_raw',      # Reallocated Sectors
    'smart_9_raw',      # Power-On Hours
    'smart_12_raw',     # Power Cycle Count
    'smart_187_raw',    # Reported Uncorrectable Errors
    'smart_188_raw',    # Command Timeout
    'smart_194_raw',    # Temperature
    'smart_197_raw',    # Current Pending Sectors
    'smart_198_raw'     # Offline Uncorrectable Sectors
]

# Get all CSV files
csv_files = [f for f in os.listdir(input_directory) if f.endswith('.csv')]

print(f"Found {len(csv_files)} CSV file(s) to process:\n")

for filename in csv_files:
    try:
        input_file = os.path.join(input_directory, filename)
        output_file = os.path.join(output_directory, filename)
        
        print(f"Processing: {filename}...")
        
        # Read only the columns you need
        df = pd.read_csv(input_file, usecols=columns_to_keep)
        
        # Save to _filtered directory
        df.to_csv(output_file, index=False)
        
        print(f"  ✓ Saved to: _filtered/{filename}")
        print(f"  ✓ Columns: {len(df.columns)}, Rows: {len(df):,}\n")
        
    except Exception as e:
        print(f"  ✗ Error processing {filename}: {str(e)}\n")

AI Usage Transparency

AI tools (ChatGPT/Claude) were used in the following ways during this project:

Sources

S.M.A.R.T. Attributes - NTFS.com