We record the status of each model’s benchmark as a sample on the blockchain and characterize the population for each model (or by manufacturer). By referencing the obtained statistics, if we observe a significant deviation, we can confidently issue a ‘preemptive’ ‘replacement recommendation.’ This mechanism will be established using the SORA blockchain.
各モデルのベンチマークに関する状態を標本にしてブロックチェーンに記録し各モデル別(またはメーカ別)に母集団の特徴付けを実施いたします。そうして得られた統計を参照しそこから大幅な乖離がみられる場合には確信を持って「事前」に「交換推薦」を出すことができる仕組みをSORA blockchainで確立いたします。

Transcend 240GB SSD

3 / 5
Operational Stability
2 / 5
Cost Efficiency
3 / 5
Durability (Normal Use)
3 / 5
Durability (Increased Write Cycles)
3 / 5
Overall Rating
14 / 25

Benchmark Test 1

Benchmark Test 2

Benchmark Test 3

Benchmark Test 4

Benchmark Test 5

Benchmark Test 6

Benchmark Test 7

Benchmark Test 8

Benchmark Test 9

Benchmark Test 10

This is a benchmark test. After conducting multiple consecutive tests, we detected a phenomenon where the deviation rate in the red-circled Test D significantly increased only on even-numbered tests. Furthermore, as this pattern was consistent over 10 tests, we concluded that it’s not a coincidence.
複数回を続けて実施いたしました結果、偶数回に限って赤丸部分のテストDの乖離率が大幅に上昇する現象を検出いたしました。 また、計10回実施いたしましてきちっと5回揃いましたので、偶然ではないと判断いたしました。

Test D involves random write operations at the smallest scale of 4KB, benchmarked under a load varying from -3σ to +3σ. It seems like an algorithm that accumulates tasks during the -1σ to -3σ phase and then releases them all at once. During this release, the weighted average drops significantly, which is reflected in the deviation rate. The ‘alternating occurrence’ of this phenomenon is particularly intriguing.

The duration of each benchmark test is roughly a few minutes, suggesting that the system can withstand the load during this time and then releases it all afterward. This implies a focus on benchmark performance.
ベンチマークの測定時間は、だいたい数分ゆえに、その時間は何とか持ちこたえて、まとめて後から放出・・という感じでしょうか。 つまり、ベンチマーク重視のような感じです。

However, this significant drop in the weighted average can lead to logical failures, so a more distributed release would be preferable. Although distributed release might cause disruptions during the benchmark (like a messy graph), it has no practical impact on actual use.