A proposed model called REE-TM for managing heterogeneous cloud traffic reported higher reliability and lower energy use than two reduced versions in simulations. Resource utilization was reported as 1% to 4% lower than an OPTIMAL comparator. The evaluation was simulation-based, and the supplied text does not report a randomized experiment or live deployment.
Across the simulated settings, REE-TM reliability ranged from 85.6% to 97.8%, resource utilization from 42.6% to 49.9%, and overloads from 1.95% to 8.20%. Load-distribution success reached 99.9%.
The reported comparison
Compared with the two reduced versions, the model’s reported reliability was up to 26.04% higher than W-REE-TM∗ and up to 30.25% higher than W-REE-TM∗∗.
Resource utilization was reported up to 22.35% higher than W-REE-TM∗ and up to 39.81% higher than W-REE-TM∗∗, respectively, while it was 1% to 4% lower than OPTIMAL.
Energy use was reported to be 20% to 23% lower than the two W-REE-TM baselines and 1% to 5% higher than OPTIMAL.
At 1,000 requests, failure occurrences were reported up to 66.67% lower than W-REE-TM∗ and 87.9% lower than W-REE-TM∗∗. Those reductions were based on upper-quartile values, and no uncertainty intervals were reported for that comparison.
At the same request count, active physical-node counts were reported up to 11.11% lower than W-REE-TM∗ and 15.79% lower than W-REE-TM∗∗.
What REE-TM combines
REE-TM groups tasks into Stragglers, Resource hogs and Normal jobs. It uses QML-WE with TG-QNN for workload estimation and TSECE for traffic-entropy analysis.
For workload prediction, TG-QNN had an MSE (mean squared error) of 0.0005 and an MAE (mean absolute error) of 0.0099 at a five-minute prediction window. At 60 minutes, the reported MSE was 0.0024 and the MAE was 0.0336. At 10 minutes, ATTN slightly outperformed TG-QNN in the reported comparison.
Among the compared optimization methods, QBHO had the highest reported average accuracy, at 98.06%, and the lowest reported space complexity, a measure of memory demand. Its reported time complexity was on the order of O(Zn2 M K).
The trace and setup
The test used the Google Compute Cluster dataset, which contains information on CPU, memory, disk I/O and resource use from 672,300 jobs executed on 12,500 servers over 29 days.
The simulation setup specified two Intel Xeon Silver 4114 CPUs, 40 cores, a 2.20 GHz clock speed and 128 GB of memory. It represented three physical-server types and four virtual-node configurations.
A component-by-component sequence reported reliability of 66.67% and success of 80.98% for TGQNN+TM. With QBHO added, the corresponding figures were 73.78% and 88.98%; with TSECE included as well, they were 86.88% and 98.6%.
Where the evidence ends
The supplied evaluation is simulation-based, and no randomized experiment or live deployment is reported. The comparisons should therefore be read as results from the evaluated simulation settings rather than as a live-cloud report.
No confidence intervals were reported for the reliability, resource-utilization, energy, failure or active-node comparisons. The ablation sequence also came without confidence intervals or statistical tests.
The front matter states that the article was accepted in IEEE Transactions on Cloud Computing and is © 2025 IEEE. No study-specific funding statement is provided; the author biography states that Deepika Saxena is a recipient of the JSPS KAKENHI Early Career Young Scientist Research Grant.
Paper data and sources
Original title: REE-TM: Reliable and Energy-Efficient Traffic Management Model for Diverse Cloud Workloads
Authors: Ashutosh Kumar Singh, Deepika Saxena, Volker Lindenstruth
Journal/Repository: IEEE Transactions on Cloud Computing 2025
Status: Peer-reviewed
First online: 2026-08-26
DOI: 10.1109/tcc.2025.3581697
Original paper · Full text