|
| 1 | +# DAU search and landing-page plan |
| 2 | + |
| 3 | +This plan keeps commercial search content evidence-led. It does not propose |
| 4 | +publishing thin comparison pages or performance claims before reproducible |
| 5 | +measurements exist. |
| 6 | + |
| 7 | +## Position |
| 8 | + |
| 9 | +DAU is workload-shaped FPGA acceleration for analytical software. It selectively |
| 10 | +maps supported Polars, dataframe, time-series, and domain operations into |
| 11 | +right-sized streaming hardware while unsupported work remains in familiar |
| 12 | +software. |
| 13 | + |
| 14 | +Primary distinction: |
| 15 | + |
| 16 | +- not a fixed database appliance; |
| 17 | +- not per-query synthesis in the runtime path; |
| 18 | +- not a claim that every dataframe operation belongs on an FPGA; |
| 19 | +- a reusable platform whose configurations improve around recurring workloads |
| 20 | + and scale across device classes. |
| 21 | + |
| 22 | +## Search-intent clusters |
| 23 | + |
| 24 | +Use these terms naturally in useful pages. Do not add a `meta keywords` tag or |
| 25 | +repeat phrases solely for ranking. |
| 26 | + |
| 27 | +### FPGA analytical acceleration |
| 28 | + |
| 29 | +- FPGA analytics accelerator |
| 30 | +- FPGA database acceleration |
| 31 | +- hardware-accelerated dataframes |
| 32 | +- reconfigurable dataflow analytics |
| 33 | +- FPGA query acceleration |
| 34 | +- composable FPGA accelerator |
| 35 | + |
| 36 | +### Polars and dataframe acceleration |
| 37 | + |
| 38 | +- Polars FPGA acceleration |
| 39 | +- Polars hardware acceleration |
| 40 | +- accelerate Polars queries |
| 41 | +- Polars GPU alternative |
| 42 | +- dataframe hardware acceleration |
| 43 | +- CPU GPU FPGA dataframe comparison |
| 44 | + |
| 45 | +### Time-series and market data |
| 46 | + |
| 47 | +- FPGA time-series analytics |
| 48 | +- market-data analytics acceleration |
| 49 | +- hardware-accelerated OHLCV |
| 50 | +- time-series aggregation FPGA |
| 51 | +- as-of join acceleration |
| 52 | +- resident market-data analytics |
| 53 | + |
| 54 | +Publish terms tied to as-of joins, rolling analytics, or options only after the |
| 55 | +named operation is implemented and measured. |
| 56 | + |
| 57 | +## Recommended pages |
| 58 | + |
| 59 | +### Workload pages |
| 60 | + |
| 61 | +1. `/workloads/time-series-analytics` |
| 62 | + - Public target after TSBS coverage is frozen. |
| 63 | + - Explain bars, windows, rates, resampling, state bounds, and residency. |
| 64 | +2. `/workloads/market-data` |
| 65 | + - Lead with the measured OHLCV workflow. |
| 66 | + - Add quote/trade as-of fusion only after silicon validation. |
| 67 | +3. `/workloads/polars-dataframes` |
| 68 | + - Explain lazy-plan capture, selective offload, and explicit fallback in |
| 69 | + user language. |
| 70 | +4. `/workloads/options-pricing` |
| 71 | + - Hold until analytic pricing or PDE hardware exists and has accuracy and |
| 72 | + native CPU baselines. |
| 73 | + |
| 74 | +### Comparison and educational pages |
| 75 | + |
| 76 | +1. `/compare/polars-cpu` |
| 77 | + - First comparison page because a measured baseline already exists. |
| 78 | + - Show cold, resident, and break-even behavior rather than one speedup. |
| 79 | +2. `/compare/gpu-vs-fpga-analytics` |
| 80 | + - Educational decision guide covering workload shape, transfer, residency, |
| 81 | + precision, power, programmability, and deployment. |
| 82 | +3. `/compare/nvidia-rapids-cudf` |
| 83 | + - Publish only after running the same predeclared Polars workflows on a |
| 84 | + documented NVIDIA system. |
| 85 | + - Acknowledge cuDF's broad GPU operator coverage and Polars integration. |
| 86 | +4. `/compare/amd-vitis-database` |
| 87 | + - Compare platform models and developer workflow first. |
| 88 | + - Do not make performance claims without matched hardware and queries. |
| 89 | +5. `/compare/fixed-accelerator-vs-workload-shaped` |
| 90 | + - Vendor-neutral explanation of DAU's flywheel, cache, and right-sizing. |
| 91 | + |
| 92 | +Do not target DuckDB, ClickHouse, or cloud warehouses as direct competitors |
| 93 | +until DAU exposes a comparable complete database surface. They can appear in |
| 94 | +CPU-baseline methodology without implying product equivalence. |
| 95 | + |
| 96 | +## Page template |
| 97 | + |
| 98 | +Every workload or comparison page should contain: |
| 99 | + |
| 100 | +1. a unique descriptive title and one H1; |
| 101 | +2. a direct answer to the searcher's decision; |
| 102 | +3. the workload and system boundaries; |
| 103 | +4. what DAU accelerates and what remains in software; |
| 104 | +5. a reproducible benchmark or an explicit “not yet measured” statement; |
| 105 | +6. cold, resident, and end-to-end costs where applicable; |
| 106 | +7. “Choose DAU when…” and “Choose the alternative when…”; |
| 107 | +8. limitations and current product status; |
| 108 | +9. links to relevant platform, proof, developer, and design-partner content; |
| 109 | +10. page-specific Open Graph metadata, canonical URL, and sitemap entry. |
| 110 | + |
| 111 | +Add FAQ structured data only when the same questions and answers are visible on |
| 112 | +the page. Add Product structured data only when DAU has a purchasable product |
| 113 | +with accurate availability and offer information. |
| 114 | + |
| 115 | +## Evidence gate |
| 116 | + |
| 117 | +A named comparison requires: |
| 118 | + |
| 119 | +- frozen workload and dataset; |
| 120 | +- documented CPU/GPU/FPGA system configurations; |
| 121 | +- competent native baseline rather than a Python row UDF; |
| 122 | +- correctness or numerical-accuracy criteria; |
| 123 | +- complete host preparation, transfer, execution, and result time; |
| 124 | +- repeated runs and distributions; |
| 125 | +- publication of unfavorable results and crossover regions; |
| 126 | +- copy reviewed against the source vendor's current documentation. |
| 127 | + |
| 128 | +## Site mechanics |
| 129 | + |
| 130 | +- Keep canonical URLs absolute and include only canonical pages in the sitemap. |
| 131 | +- Give each page a distinct title and meta description that matches visible |
| 132 | + content. |
| 133 | +- Use descriptive internal links rather than “learn more.” |
| 134 | +- Generate a 1200 × 630 social card for each major landing page. |
| 135 | +- Keep Organization and WebSite structured data on the home page; add |
| 136 | + BreadcrumbList to nested pages. |
| 137 | +- Submit the root sitemap through Google Search Console after deployment. |
| 138 | +- Validate structured data with Google's Rich Results Test and social previews |
| 139 | + with the platform debuggers after every metadata or image change. |
| 140 | + |
| 141 | +## Initial sequence |
| 142 | + |
| 143 | +1. Home page SEO and social preview. |
| 144 | +2. Measured market-data workload page. |
| 145 | +3. Polars/dataframe integration page. |
| 146 | +4. Polars CPU comparison with cold/resident break-even curves. |
| 147 | +5. Vendor-neutral GPU-versus-FPGA guide. |
| 148 | +6. Named cuDF and Vitis comparisons only after matched evaluations. |
0 commit comments