UltraEP is the first exact-load real-time expert balancer for large-EP MoE training and serving on rack-scale nodes, reaching 94.3% of ideal throughput and 1.49x over no-balancing.
Improving network availability with protective reroute
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
representative citing papers
Analysis of 50 RPKI RFCs using differential fuzzing and Internet crawling identifies 61 inconsistencies in implementations, traces 23 to spec flaws, and finds two new CVEs.
OnlineTE uses optimization decomposition to enable distributed, near-optimal traffic engineering that reacts in seconds to changes in large WANs and outperforms prior centralized approaches in emulation.
UIFO is a new scheduling model that adds class-level dynamic reordering to programmable hardware while generalizing PIFO and PIEO and sustaining 100 Gbps line rate.
Sublime generalizes Count-Min and Count Sketch with dynamically elongating counters and expanding counter arrays to deliver sublinear error growth and lower memory use on skewed unbounded streams.
DODOCO measurements show MoE routing imbalance is intrinsic to architecture and real text, not correctable by EP scaling or represented by mock tokens, forming two persistent Gini bands.
Gemini Ultra reaches human-expert performance on MMLU for the first time and sets new state-of-the-art results on 30 of 32 benchmarks, including all 20 multimodal ones tested.
Two-year empirical study of 472 IXPs finds 49.2% global traffic growth, stable utilization rates, regionally distinct patterns, and high self-similarity, establishing IXP statistics as a robust proxy for overall Internet dynamics.
citing papers explorer
-
UltraEP: Unleash MoE Training and Inference on Rack-Scale Nodes with Near-Optimal Load Balancing
UltraEP is the first exact-load real-time expert balancer for large-EP MoE training and serving on rack-scale nodes, reaching 94.3% of ideal throughput and 1.49x over no-balancing.
-
The Fault in Our Drafts: Vulnerabilities in RPKI Specification and Software
Analysis of 50 RPKI RFCs using differential fuzzing and Internet crawling identifies 61 inconsistencies in implementations, traces 23 to spec flaws, and finds two new CVEs.
-
Near-optimal Online Traffic Engineering
OnlineTE uses optimization decomposition to enable distributed, near-optimal traffic engineering that reacts in seconds to changes in large WANs and outperforms prior centralized approaches in emulation.
-
Programmable Packet Scheduling with Dynamic Reordering at Line Rate
UIFO is a new scheduling model that adds class-level dynamic reordering to programmable hardware while generalizing PIFO and PIEO and sustaining 100 Gbps line rate.
-
Sublime: Sublinear Error & Space for Unbounded Skewed Streams
Sublime generalizes Count-Min and Count Sketch with dynamically elongating counters and expanding counter arrays to deliver sublinear error growth and lower memory use on skewed unbounded streams.
-
Diagnosing Overhead in Dispatch Operations: Cross-architecture Observatory
DODOCO measurements show MoE routing imbalance is intrinsic to architecture and real text, not correctable by EP scaling or represented by mock tokens, forming two persistent Gini bands.
-
Five Blind Men and the Internet: Towards an Understanding of Internet Traffic
Two-year empirical study of 472 IXPs finds 49.2% global traffic growth, stable utilization rates, regionally distinct patterns, and high self-similarity, establishing IXP statistics as a robust proxy for overall Internet dynamics.