[jira] [Updated] (SPARK-24615) Accelerator-aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Xingbo Jiang updated SPARK-24615: - Epic Name: Support GPU Scheduling > Accelerator-aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Epic > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Assignee: Xingbo Jiang >Priority: Major > Labels: Hydrogen, SPIP > Attachments: Accelerator-aware scheduling in Apache Spark 3.0.pdf, > SPIP_ Accelerator-aware scheduling.pdf > > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. It doesn't cover all the implementation > details, just highlight the parts should be changed. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-24615) Accelerator-aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Xingbo Jiang updated SPARK-24615: - Issue Type: Epic (was: Improvement) > Accelerator-aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Epic > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Assignee: Xingbo Jiang >Priority: Major > Labels: Hydrogen, SPIP > Attachments: Accelerator-aware scheduling in Apache Spark 3.0.pdf, > SPIP_ Accelerator-aware scheduling.pdf > > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. It doesn't cover all the implementation > details, just highlight the parts should be changed. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-24615) Accelerator-aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Xiangrui Meng updated SPARK-24615: -- Attachment: SPIP_ Accelerator-aware scheduling.pdf > Accelerator-aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Improvement > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Assignee: Xingbo Jiang >Priority: Major > Labels: Hydrogen, SPIP > Attachments: Accelerator-aware scheduling in Apache Spark 3.0.pdf, > SPIP_ Accelerator-aware scheduling.pdf > > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. It doesn't cover all the implementation > details, just highlight the parts should be changed. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-24615) Accelerator-aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Xiangrui Meng updated SPARK-24615: -- Attachment: Accelerator-aware scheduling in Apache Spark 3.0.pdf > Accelerator-aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Improvement > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Assignee: Xingbo Jiang >Priority: Major > Labels: Hydrogen, SPIP > Attachments: Accelerator-aware scheduling in Apache Spark 3.0.pdf, > SPIP_ Accelerator-aware scheduling.pdf > > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. It doesn't cover all the implementation > details, just highlight the parts should be changed. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-24615) Accelerator-aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Xiangrui Meng updated SPARK-24615: -- Summary: Accelerator-aware task scheduling for Spark (was: Accelerator aware task scheduling for Spark) > Accelerator-aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Improvement > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Assignee: Saisai Shao >Priority: Major > Labels: Hydrogen, SPIP > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. It doesn't cover all the implementation > details, just highlight the parts should be changed. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-24615) Accelerator aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Xiangrui Meng updated SPARK-24615: -- Shepherd: Xiangrui Meng > Accelerator aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Improvement > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Priority: Major > Labels: Hydrogen, SPIP > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. It doesn't cover all the implementation > details, just highlight the parts should be changed. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-24615) Accelerator aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Xiangrui Meng updated SPARK-24615: -- Labels: Hydrogen SPIP (was: SPIP) > Accelerator aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Improvement > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Priority: Major > Labels: Hydrogen, SPIP > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. It doesn't cover all the implementation > details, just highlight the parts should be changed. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-24615) Accelerator aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Saisai Shao updated SPARK-24615: Labels: SPIP (was: ) > Accelerator aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Improvement > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Priority: Major > Labels: SPIP > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. It doesn't cover all the implementation > details, just highlight the parts should be changed. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-24615) Accelerator aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Saisai Shao updated SPARK-24615: Description: In the machine learning area, accelerator card (GPU, FPGA, TPU) is predominant compared to CPUs. To make the current Spark architecture to work with accelerator cards, Spark itself should understand the existence of accelerators and know how to schedule task onto the executors where accelerators are equipped. Current Spark’s scheduler schedules tasks based on the locality of the data plus the available of CPUs. This will introduce some problems when scheduling tasks with accelerators required. # CPU cores are usually more than accelerators on one node, using CPU cores to schedule accelerator required tasks will introduce the mismatch. # In one cluster, we always assume that CPU is equipped in each node, but this is not true of accelerator cards. # The existence of heterogeneous tasks (accelerator required or not) requires scheduler to schedule tasks with a smart way. So here propose to improve the current scheduler to support heterogeneous tasks (accelerator requires or not). This can be part of the work of Project hydrogen. Details is attached in google doc. It doesn't cover all the implementation details, just highlight the parts should be changed. CC [~yanboliang] [~merlintang] was: In the machine learning area, accelerator card (GPU, FPGA, TPU) is predominant compared to CPUs. To make the current Spark architecture to work with accelerator cards, Spark itself should understand the existence of accelerators and know how to schedule task onto the executors where accelerators are equipped. Current Spark’s scheduler schedules tasks based on the locality of the data plus the available of CPUs. This will introduce some problems when scheduling tasks with accelerators required. # CPU cores are usually more than accelerators on one node, using CPU cores to schedule accelerator required tasks will introduce the mismatch. # In one cluster, we always assume that CPU is equipped in each node, but this is not true of accelerator cards. # The existence of heterogeneous tasks (accelerator required or not) requires scheduler to schedule tasks with a smart way. So here propose to improve the current scheduler to support heterogeneous tasks (accelerator requires or not). This can be part of the work of Project hydrogen. Details is attached in google doc. CC [~yanboliang] [~merlintang] > Accelerator aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Improvement > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Priority: Major > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. It doesn't cover all the implementation > details, just highlight the parts should be changed. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Updated] (SPARK-24615) Accelerator aware task scheduling for Spark
[ https://issues.apache.org/jira/browse/SPARK-24615?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Saisai Shao updated SPARK-24615: Description: In the machine learning area, accelerator card (GPU, FPGA, TPU) is predominant compared to CPUs. To make the current Spark architecture to work with accelerator cards, Spark itself should understand the existence of accelerators and know how to schedule task onto the executors where accelerators are equipped. Current Spark’s scheduler schedules tasks based on the locality of the data plus the available of CPUs. This will introduce some problems when scheduling tasks with accelerators required. # CPU cores are usually more than accelerators on one node, using CPU cores to schedule accelerator required tasks will introduce the mismatch. # In one cluster, we always assume that CPU is equipped in each node, but this is not true of accelerator cards. # The existence of heterogeneous tasks (accelerator required or not) requires scheduler to schedule tasks with a smart way. So here propose to improve the current scheduler to support heterogeneous tasks (accelerator requires or not). This can be part of the work of Project hydrogen. Details is attached in google doc. CC [~yanboliang] [~merlintang] was: In the machine learning area, accelerator card (GPU, FPGA, TPU) is predominant compared to CPUs. To make the current Spark architecture to work with accelerator cards, Spark itself should understand the existence of accelerators and know how to schedule task onto the executors where accelerators are equipped. Current Spark’s scheduler schedules tasks based on the locality of the data plus the available of CPUs. This will introduce some problems when scheduling tasks with accelerators required. # CPU cores are usually more than accelerators on one node, using CPU cores to schedule accelerator required tasks will introduce the mismatch. # In one cluster, we always assume that CPU is equipped in each node, but this is not true of accelerator cards. # The existence of heterogeneous tasks (accelerator required or not) requires scheduler to schedule tasks with a smart way. So here propose to improve the current scheduler to support heterogeneous tasks (accelerator requires or not). Details is attached in google doc. > Accelerator aware task scheduling for Spark > --- > > Key: SPARK-24615 > URL: https://issues.apache.org/jira/browse/SPARK-24615 > Project: Spark > Issue Type: Improvement > Components: Spark Core >Affects Versions: 2.4.0 >Reporter: Saisai Shao >Priority: Major > > In the machine learning area, accelerator card (GPU, FPGA, TPU) is > predominant compared to CPUs. To make the current Spark architecture to work > with accelerator cards, Spark itself should understand the existence of > accelerators and know how to schedule task onto the executors where > accelerators are equipped. > Current Spark’s scheduler schedules tasks based on the locality of the data > plus the available of CPUs. This will introduce some problems when scheduling > tasks with accelerators required. > # CPU cores are usually more than accelerators on one node, using CPU cores > to schedule accelerator required tasks will introduce the mismatch. > # In one cluster, we always assume that CPU is equipped in each node, but > this is not true of accelerator cards. > # The existence of heterogeneous tasks (accelerator required or not) > requires scheduler to schedule tasks with a smart way. > So here propose to improve the current scheduler to support heterogeneous > tasks (accelerator requires or not). This can be part of the work of Project > hydrogen. > Details is attached in google doc. > > CC [~yanboliang] [~merlintang] -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
