π₯ A curated list of DL cluster scheduling papers.
Please feel free to pull requests or open an issue to add papers.
| Scheduler | Year | Series | Paper | Objective | Heter. | Elastic | AutoML | Code |
|---|---|---|---|---|---|---|---|---|
| Acme | 2024 | NSDI | Paper | β£ | - | - | - | Code |
| Cassini | 2024 | NSDI | Paper | β β£β₯ | - | β | - | - |
| Sia | 2023 | SOSP | Paper | β β£β₯ | β | β | β | Code |
| EasyScale | 2023 | SC | Paper | β β£ | β | β | - | Code |
| Hydro | 2023 | OSDI | Paper | β β² | β | β | β | Code |
| Shockwave | 2023 | NSDI | Paper | β β£β₯ | - | β | - | Code |
| ModelKeeper | 2023 | NSDI | Paper | β¦β£ | - | - | β | Code |
| Lyra | 2023 | EuroSys | Paper | β β£ | β | β | - | - |
| SiloD | 2023 | EuroSys | Paper | β β£β₯ | β* | - | - | - |
| FGD | 2023 | ATC | Paper | β β£ | - | - | - | Code |
| ElasticFlow | 2023 | ASPLOS | Paper | β£βΏ | - | β | - | Code |
| Lucid | 2023 | ASPLOS | Paper | β β£ | - | - | - | Code |
| PowerFlow | 2023 | arxiv | Paper | β¦β£ | - | β | - | - |
| EDL | 2022 | TPDS | Paper | β β£ | - | β | - | - |
| AOnline | 2022 | TCC | Paper | β β£ | - | β | - | - |
| Titan | 2022 | SoCC | Paper | β β£ | - | - | - | - |
| Muri | 2022 | SIGCOMM | Paper | β β£ | β* | - | - | Code |
| Synergy | 2022 | OSDI | Paper | β£ | - | - | - | Code |
| Ali-MLaaS | 2022 | NSDI | Paper | β β£ | - | - | - | Code |
| GADGET | 2022 | INFOCOM | Paper | β β£ | - | β | - | Code |
| CloudBrain | 2022 | ICCD | Paper | β | - | - | - | Code |
| Aryl | 2022 | arxiv | Paper | β β£ | - | β | β | - |
| Singularity | 2022 | arxiv | Paper | β β£β¦ | - | β | - | - |
| $DL^2$ | 2021 | TPDS | Paper | β£ | - | β | - | Code |
| Astraea | 2021 | TPDS | Paper | β₯ | - | - | - | Code |
| Horus | 2021 | TPDS | Paper | β β£ | - | - | - | - |
| Liquid | 2021 | TPDS | Paper | β£ | - | - | - | Code |
| POP | 2021 | SOSP | Paper | β₯β£ | β | - | - | Code |
| Chronus | 2021 | SoCC | Paper | βΏ | - | - | - | Code |
| SEER | 2021 | SoCC | Paper | β² | - | β | β | - |
| Helios | 2021 | SC | Paper | β£β¦ | - | - | - | Code |
| ONES | 2021 | SC | Paper | β β£ | - | β | - | Code |
| Pollux | 2021 | OSDI | Paper | β β£β₯ | - | β | β | Code |
| AFS | 2021 | NSDI | Paper | β β£ | - | β | - | - |
| SMD | 2021 | INFOCOM | Paper | β£ | - | - | - | - |
| ANDREAS | 2021 | FCloud | Paper | β¦ | - | - | - | - |
| RubberBand | 2021 | EuroSys | Paper | β¦ | - | β | β | - |
| Hermes | 2021 | Electronics | Paper | β£ | - | - | β | - |
| Jigsaw | 2021 | DistributedML | Paper | β£ | - | - | - | - |
| DynamoML | 2021 | CLOSER | Paper | β β£ | - | β | - | - |
| GENIE | 2020 | TPDS | Paper | βΏ | - | β | - | - |
| Parrot | 2020 | TCC | Paper | β£ | - | - | - | - |
| Non-Intrusive | 2020 | SC | Paper | β β£ | - | β | - | - |
| Antman | 2020 | OSDI | Paper | β β£ | - | β | - | Code |
| Gavel | 2020 | OSDI | Paper | β£β₯ | β | - | - | Code |
| HiveD | 2020 | OSDI | Paper | β£ | - | - | - | Code |
| Themis | 2020 | NSDI | Paper | β₯ | - | - | - | - |
| Salus | 2020 | MLSys | Paper | β β£ | - | - | - | Code |
| Vaibhav et al. | 2020 | MASCOTS | Paper | β β£ | - | β | - | - |
| SPIN | 2020 | INFOCOM | Paper | β£ | - | - | - | - |
| E-LAS | 2020 | ICPP | Paper | β£ | - | - | - | - |
| CODA | 2020 | ICDCS | Paper | β£ | β* | - | - | - |
| Elan | 2020 | ICDCS | Paper | β β£ | - | β | - | - |
| Yeung | 2020 | HotCloud | Paper | β | - | - | - | - |
| $Gandiva_{fair}$ | 2020 | EuroSys | Paper | β₯β£ | β | - | - | - |
| MLCloudPrice | 2020 | DISPA | Paper | β£β¦ | - | - | - | Code |
| MLFS | 2020 | CoNext | Paper | β£βΏ | - | - | - | Code |
| MARBLE | 2020 | CCGRID | Paper | β β£ | - | β | - | - |
| Ada-SRSF | 2020 | arxiv | Paper | β£ | β* | - | - | - |
| Co-scheML | 2020 | ACSOS | Paper | β£ | - | - | - | - |
| HyperSched | 2019 | SoCC | Paper | βΏ β² | - | β | β | - |
| Tiresias | 2019 | NSDI | Paper | β£ | - | - | - | Code |
| FfDL | 2019 | Middleware | Paper | β£ | - | - | - | Code |
| JPAS | 2019 | JNCA | Paper | β£β² | - | - | β | - |
| Harmony | 2019 | INFOCOM | Paper | β£ | - | - | - | - |
| Cynthia | 2019 | ICPP | Paper | β¦ | - | β | - | - |
| Jahani | 2019 | ICCCS | Paper | β¦ | β | β | - | - |
| $Sched^2$ | 2019 | GLOBECOM | Paper | β£ | - | - | - | - |
| Dragon | 2019 | CLOSER | Paper | β β£ | - | β | - | - |
| $FC^2$ | 2019 | CC | Paper | β¦ | β* | β | - | - |
| Philly | 2019 | ATC | Paper | β£ | - | - | - | Code |
| Gandiva | 2018 | OSDI | Paper | β β£ | - | β | β | - |
| OASiS | 2018 | INFOCOM | Paper | β β£ | - | β | - | - |
| Optimus | 2018 | EuroSys | Paper | β£ | - | β | - | Code |
| Dorm | 2017 | SMARTCOMP | Paper | β₯ | - | - | - | - |
| Topology-Aware | 2017 | SC | Paper | β£ | - | - | - | Code |
| HyperDrive | 2017 | Middleware | Paper | β£β² | - | - | β | - |
Symbols of Training Schedulers:
| JCT | Utilization | Cost | Fairness | DDL | Accuracy |
|---|---|---|---|---|---|
| β£ | β | β¦ | β₯ | βΏ | β² |
| Scheduler | Year | Series | Paper | Objective | Batch | Share | Cloud | Source Code |
|---|---|---|---|---|---|---|---|---|
| SpotServe | 2024 | ASPLOS | Paper | β¦β₯ | β | - | β | Code |
| DeltaZip | 2023 | arxiv | Paper | β₯β | β | - | - | Code |
| MOSEL | 2023 | arxiv | Paper | β¦β | β | - | - | - |
| Punica | 2023 | arxiv | Paper | β₯β | - | β* | - | Code |
| S-LoRA | 2023 | arxiv | Paper | β₯β | - | β* | - | Code |
| Symphony | 2023 | arxiv | Paper | βΏβ β¦ | β | - | - | - |
| DeepPlan | 2023 | EuroSys | Paper | β¦β | β | - | - | Code |
| Tabi | 2023 | EuroSys | Paper | β¦β£ | - | - | β | - |
| Kairos | 2023 | HPDC | Paper | β¦β₯β | β | - | β | Code |
| Shepherd | 2023 | NSDI | Paper | βΏβ β¦ | β | - | - | - |
| AlpaServe | 2023 | OSDI | Paper | β¦β | β | β | - | Code |
| Clover | 2023 | SC | Paper | β₯ | β | β | - | Code |
| iGniter | 2023 | TPDS | Paper | βΏβ₯ | β | β | β | Code |
| Gpulet | 2022 | ATC | Paper | βΏβ β₯ | β | β | - | Code |
| Cocktail | 2022 | NSDI | Paper | β£β¦β₯ | - | - | β | Code |
| INFaaS | 2021 | ATC | Paper | β¦β₯β | - | β | β | Code |
| MIG-SERVING | 2021 | CoRR | Paper | β¦β₯ | β | β | - | - |
| Mendoza et al. | 2021 | EuroMLSys | Paper | β¦ | - | β | - | - |
| Abacus | 2021 | SC | Paper | β¦β | - | β | - | Code |
| Morphling | 2021 | SoCC | Paper | β₯β | β | β | β | Code |
| Irina | 2020 | APNet | Paper | β¦β βΏ | β | β | - | - |
| DyBatch | 2020 | CCGrid | Paper | β¦β | β | β | - | - |
| CMS | 2020 | Future Internet | Paper | β£βΏ | - | - | - | - |
| PERSEUS | 2020 | IC2E | Paper | β¦β₯β | β | - | β | Code |
| AutoDeep | 2020 | Infocom | Paper | β¦β₯β | - | β | β | - |
| Clockwork | 2020 | OSDI | Paper | β¦β | β | - | - | Code |
| GSLICE | 2020 | SoCC | Paper | β βΏ | β | β | - | - |
| Inferline | 2020 | SoCC | Paper | β¦β₯ | β | - | β | Code |
| MArk | 2019 | ATC | Paper | β¦β₯ | β | - | β | Code |
| TrIMS | 2019 | CLOUD | Paper | β¦β βΏ | β | β | β | Code |
| Kube-Knots | 2019 | CLUSTER | Paper | β¦βΏ | β | β | - | - |
| Gilman et al. | 2019 | DIDL | Paper | β¦β | - | β | - | - |
| Nanily | 2019 | HPCC | Paper | β¦β | β | - | - | - |
| Ebird | 2019 | ICCD | Paper | β¦β βΏ | β | β | - | Code |
| Tolerance Tiers | 2019 | ISPASS | Paper | β£β¦β₯ | - | - | β | - |
| RRL | 2019 | SC | Paper | β¦ | β | β | - | Code |
| ParM | 2019 | SOSP | Paper | β¦ | β | - | - | Code |
| HiveMind | 2018 | NIPS | Paper | β | β | β | - | - |
| Space-Time | 2018 | NIPS | Paper | β βΏ | β | β | - | - |
| Ease.ml | 2018 | VLDB | Paper | β£ | - | - | - | Code |
| Rafiki | 2018 | VLDB | Paper | β£β¦ | β | - | - | Code |
| Clipper | 2017 | NSDI | Paper | β£β¦β | β | - | - | Code |
Symbols of Inference Schedulers:
| Accuracy | Throughput | Latency | Cost | Utilization |
|---|---|---|---|---|
| β£ | β | β¦ | β₯ | βΏ |
| Terminology | Definition |
|---|---|
| JCT | Job Completion Time (Job Finish Time - Job Submission Time) |
| Fairness | a metric to assess whether resources are fairly shared among users or jobs |
| QoS | Quality of Service |
| DDL | Deadline, a time point where DL job must be completed |
| SLO | Service Level Objective |
π₯ A curated list of DL cluster scheduling papers.
Please feel free to pull requests or open an issue to add papers.
| Scheduler | Year | Series | Paper | Objective | Heter. | Elastic | AutoML | Code |
|---|---|---|---|---|---|---|---|---|
| Acme | 2024 | NSDI | Paper | β£ | - | - | - | Code |
| Cassini | 2024 | NSDI | Paper | β β£β₯ | - | β | - | - |
| Sia | 2023 | SOSP | Paper | β β£β₯ | β | β | β | Code |
| EasyScale | 2023 | SC | Paper | β β£ | β | β | - | Code |
| Hydro | 2023 | OSDI | Paper | β β² | β | β | β | Code |
| Shockwave | 2023 | NSDI | Paper | β β£β₯ | - | β | - | Code |
| ModelKeeper | 2023 | NSDI | Paper | β¦β£ | - | - | β | Code |
| Lyra | 2023 | EuroSys | Paper | β β£ | β | β | - | - |
| SiloD | 2023 | EuroSys | Paper | β β£β₯ | β* | - | - | - |
| FGD | 2023 | ATC | Paper | β β£ | - | - | - | Code |
| ElasticFlow | 2023 | ASPLOS | Paper | β£βΏ | - | β | - | Code |
| Lucid | 2023 | ASPLOS | Paper | β β£ | - | - | - | Code |
| PowerFlow | 2023 | arxiv | Paper | β¦β£ | - | β | - | - |
| EDL | 2022 | TPDS | Paper | β β£ | - | β | - | - |
| AOnline | 2022 | TCC | Paper | β β£ | - | β | - | - |
| Titan | 2022 | SoCC | Paper | β β£ | - | - | - | - |
| Muri | 2022 | SIGCOMM | Paper | β β£ | β* | - | - | Code |
| Synergy | 2022 | OSDI | Paper | β£ | - | - | - | Code |
| Ali-MLaaS | 2022 | NSDI | Paper | β β£ | - | - | - | Code |
| GADGET | 2022 | INFOCOM | Paper | β β£ | - | β | - | Code |
| CloudBrain | 2022 | ICCD | Paper | β | - | - | - | Code |
| Aryl | 2022 | arxiv | Paper | β β£ | - | β | β | - |
| Singularity | 2022 | arxiv | Paper | β β£β¦ | - | β | - | - |
| $DL^2$ | 2021 | TPDS | Paper | β£ | - | β | - | Code |
| Astraea | 2021 | TPDS | Paper | β₯ | - | - | - | Code |
| Horus | 2021 | TPDS | Paper | β β£ | - | - | - | - |
| Liquid | 2021 | TPDS | Paper | β£ | - | - | - | Code |
| POP | 2021 | SOSP | Paper | β₯β£ | β | - | - | Code |
| Chronus | 2021 | SoCC | Paper | βΏ | - | - | - | Code |
| SEER | 2021 | SoCC | Paper | β² | - | β | β | - |
| Helios | 2021 | SC | Paper | β£β¦ | - | - | - | Code |
| ONES | 2021 | SC | Paper | β β£ | - | β | - | Code |
| Pollux | 2021 | OSDI | Paper | β β£β₯ | - | β | β | Code |
| AFS | 2021 | NSDI | Paper | β β£ | - | β | - | - |
| SMD | 2021 | INFOCOM | Paper | β£ | - | - | - | - |
| ANDREAS | 2021 | FCloud | Paper | β¦ | - | - | - | - |
| RubberBand | 2021 | EuroSys | Paper | β¦ | - | β | β | - |
| Hermes | 2021 | Electronics | Paper | β£ | - | - | β | - |
| Jigsaw | 2021 | DistributedML | Paper | β£ | - | - | - | - |
| DynamoML | 2021 | CLOSER | Paper | β β£ | - | β | - | - |
| GENIE | 2020 | TPDS | Paper | βΏ | - | β | - | - |
| Parrot | 2020 | TCC | Paper | β£ | - | - | - | - |
| Non-Intrusive | 2020 | SC | Paper | β β£ | - | β | - | - |
| Antman | 2020 | OSDI | Paper | β β£ | - | β | - | Code |
| Gavel | 2020 | OSDI | Paper | β£β₯ | β | - | - | Code |
| HiveD | 2020 | OSDI | Paper | β£ | - | - | - | Code |
| Themis | 2020 | NSDI | Paper | β₯ | - | - | - | - |
| Salus | 2020 | MLSys | Paper | β β£ | - | - | - | Code |
| Vaibhav et al. | 2020 | MASCOTS | Paper | β β£ | - | β | - | - |
| SPIN | 2020 | INFOCOM | Paper | β£ | - | - | - | - |
| E-LAS | 2020 | ICPP | Paper | β£ | - | - | - | - |
| CODA | 2020 | ICDCS | Paper | β£ | β* | - | - | - |
| Elan | 2020 | ICDCS | Paper | β β£ | - | β | - | - |
| Yeung | 2020 | HotCloud | Paper | β | - | - | - | - |
| $Gandiva_{fair}$ | 2020 | EuroSys | Paper | β₯β£ | β | - | - | - |
| MLCloudPrice | 2020 | DISPA | Paper | β£β¦ | - | - | - | Code |
| MLFS | 2020 | CoNext | Paper | β£βΏ | - | - | - | Code |
| MARBLE | 2020 | CCGRID | Paper | β β£ | - | β | - | - |
| Ada-SRSF | 2020 | arxiv | Paper | β£ | β* | - | - | - |
| Co-scheML | 2020 | ACSOS | Paper | β£ | - | - | - | - |
| HyperSched | 2019 | SoCC | Paper | βΏ β² | - | β | β | - |
| Tiresias | 2019 | NSDI | Paper | β£ | - | - | - | Code |
| FfDL | 2019 | Middleware | Paper | β£ | - | - | - | Code |
| JPAS | 2019 | JNCA | Paper | β£β² | - | - | β | - |
| Harmony | 2019 | INFOCOM | Paper | β£ | - | - | - | - |
| Cynthia | 2019 | ICPP | Paper | β¦ | - | β | - | - |
| Jahani | 2019 | ICCCS | Paper | β¦ | β | β | - | - |
| $Sched^2$ | 2019 | GLOBECOM | Paper | β£ | - | - | - | - |
| Dragon | 2019 | CLOSER | Paper | β β£ | - | β | - | - |
| $FC^2$ | 2019 | CC | Paper | β¦ | β* | β | - | - |
| Philly | 2019 | ATC | Paper | β£ | - | - | - | Code |
| Gandiva | 2018 | OSDI | Paper | β β£ | - | β | β | - |
| OASiS | 2018 | INFOCOM | Paper | β β£ | - | β | - | - |
| Optimus | 2018 | EuroSys | Paper | β£ | - | β | - | Code |
| Dorm | 2017 | SMARTCOMP | Paper | β₯ | - | - | - | - |
| Topology-Aware | 2017 | SC | Paper | β£ | - | - | - | Code |
| HyperDrive | 2017 | Middleware | Paper | β£β² | - | - | β | - |
Symbols of Training Schedulers:
| JCT | Utilization | Cost | Fairness | DDL | Accuracy |
|---|---|---|---|---|---|
| β£ | β | β¦ | β₯ | βΏ | β² |
| Scheduler | Year | Series | Paper | Objective | Batch | Share | Cloud | Source Code |
|---|---|---|---|---|---|---|---|---|
| SpotServe | 2024 | ASPLOS | Paper | β¦β₯ | β | - | β | Code |
| DeltaZip | 2023 | arxiv | Paper | β₯β | β | - | - | Code |
| MOSEL | 2023 | arxiv | Paper | β¦β | β | - | - | - |
| Punica | 2023 | arxiv | Paper | β₯β | - | β* | - | Code |
| S-LoRA | 2023 | arxiv | Paper | β₯β | - | β* | - | Code |
| Symphony | 2023 | arxiv | Paper | βΏβ β¦ | β | - | - | - |
| DeepPlan | 2023 | EuroSys | Paper | β¦β | β | - | - | Code |
| Tabi | 2023 | EuroSys | Paper | β¦β£ | - | - | β | - |
| Kairos | 2023 | HPDC | Paper | β¦β₯β | β | - | β | Code |
| Shepherd | 2023 | NSDI | Paper | βΏβ β¦ | β | - | - | - |
| AlpaServe | 2023 | OSDI | Paper | β¦β | β | β | - | Code |
| Clover | 2023 | SC | Paper | β₯ | β | β | - | Code |
| iGniter | 2023 | TPDS | Paper | βΏβ₯ | β | β | β | Code |
| Gpulet | 2022 | ATC | Paper | βΏβ β₯ | β | β | - | Code |
| Cocktail | 2022 | NSDI | Paper | β£β¦β₯ | - | - | β | Code |
| INFaaS | 2021 | ATC | Paper | β¦β₯β | - | β | β | Code |
| MIG-SERVING | 2021 | CoRR | Paper | β¦β₯ | β | β | - | - |
| Mendoza et al. | 2021 | EuroMLSys | Paper | β¦ | - | β | - | - |
| Abacus | 2021 | SC | Paper | β¦β | - | β | - | Code |
| Morphling | 2021 | SoCC | Paper | β₯β | β | β | β | Code |
| Irina | 2020 | APNet | Paper | β¦β βΏ | β | β | - | - |
| DyBatch | 2020 | CCGrid | Paper | β¦β | β | β | - | - |
| CMS | 2020 | Future Internet | Paper | β£βΏ | - | - | - | - |
| PERSEUS | 2020 | IC2E | Paper | β¦β₯β | β | - | β | Code |
| AutoDeep | 2020 | Infocom | Paper | β¦β₯β | - | β | β | - |
| Clockwork | 2020 | OSDI | Paper | β¦β | β | - | - | Code |
| GSLICE | 2020 | SoCC | Paper | β βΏ | β | β | - | - |
| Inferline | 2020 | SoCC | Paper | β¦β₯ | β | - | β | Code |
| MArk | 2019 | ATC | Paper | β¦β₯ | β | - | β | Code |
| TrIMS | 2019 | CLOUD | Paper | β¦β βΏ | β | β | β | Code |
| Kube-Knots | 2019 | CLUSTER | Paper | β¦βΏ | β | β | - | - |
| Gilman et al. | 2019 | DIDL | Paper | β¦β | - | β | - | - |
| Nanily | 2019 | HPCC | Paper | β¦β | β | - | - | - |
| Ebird | 2019 | ICCD | Paper | β¦β βΏ | β | β | - | Code |
| Tolerance Tiers | 2019 | ISPASS | Paper | β£β¦β₯ | - | - | β | - |
| RRL | 2019 | SC | Paper | β¦ | β | β | - | Code |
| ParM | 2019 | SOSP | Paper | β¦ | β | - | - | Code |
| HiveMind | 2018 | NIPS | Paper | β | β | β | - | - |
| Space-Time | 2018 | NIPS | Paper | β βΏ | β | β | - | - |
| Ease.ml | 2018 | VLDB | Paper | β£ | - | - | - | Code |
| Rafiki | 2018 | VLDB | Paper | β£β¦ | β | - | - | Code |
| Clipper | 2017 | NSDI | Paper | β£β¦β | β | - | - | Code |
Symbols of Inference Schedulers:
| Accuracy | Throughput | Latency | Cost | Utilization |
|---|---|---|---|---|
| β£ | β | β¦ | β₯ | βΏ |
| Terminology | Definition |
|---|---|
| JCT | Job Completion Time (Job Finish Time - Job Submission Time) |
| Fairness | a metric to assess whether resources are fairly shared among users or jobs |
| QoS | Quality of Service |
| DDL | Deadline, a time point where DL job must be completed |
| SLO | Service Level Objective |