Impact of Optical Communication Link Error on Large-Scale AI Training in Data Centers
Bibliographic record
Abstract
Recently, data centers (DCs) have been increasingly dedicated to artificial intelligence (AI) training processes. To enable large-scale model training, parallelization schemes such as distributed data parallelism (DDP) and pipeline parallelism (PP) are essential. Both techniques require extensive data transmission through optical communication links within a DC. The latency and power consumption of these links are critical factors affecting DC efficiency for AI training. Although optical links are typically engineered for near-error-free transmission, the impact of data transmission errors during AI training remains insufficiently explored. In this work, we analytically investigate the effect of communication link errors on the learning process under the DDP and PP schemes. Our analysis reveals that link errors introduce bounded noise into the model weights, allowing weight error levels to be controlled by maintaining an appropriate link bit error rate (BER). Relaxing the error-free requirement opens new opportunities for optimizing optical link performance. Specifically, we propose a novel protocol stack layer for optical links, with minimal deviation from the IEEE 802.3bs standard, to enable data transmission with reduced latency. We validate our theoretical findings through simulations by training the ChatGPT2 model, consisting of 124 million parameters. The results highlight several practical implications and confirm the theoretical relationship between link BER and bounded weight noise. For example, our simulations demonstrate that a communication link with a BER less than 1e-4 has negligible impacts on DDP performance, while the PP method requires a link BER less than 1e-5.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.011 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".