Hardening Hadoop for the Enterprise Managing Diverse Workloads

  • Slides: 31
Download presentation
Hardening Hadoop for the Enterprise: Managing Diverse Workloads, Securing and Governing your Big Data

Hardening Hadoop for the Enterprise: Managing Diverse Workloads, Securing and Governing your Big Data Platform How does IT balance the tension between “one glorious cluster that serves them all” and “one cluster, one purpose – dedicated for the particular task and not to be interfered with by anything”. If they are to contain cluster sprawl, folks need help allocating a mixed workload across a shared cluster (beyond the job tracker assigning map and reduce slots), and they want to be sure the cluster is as secure as can be. Kerberos, C-groups and YARN to the rescue! This talk describes the current practices and speculates how things get better under YARN.

Agenda 1. Basics 2. • • • Cluster Evolution Vanilla Cluster Foreign Workload Introduced

Agenda 1. Basics 2. • • • Cluster Evolution Vanilla Cluster Foreign Workload Introduced Node Specialization Cluster Specialization Datacenter Integration 3. YARN 4. Security

Hadoop – and her 2 beautiful things Data I will spread your data out

Hadoop – and her 2 beautiful things Data I will spread your data out over many servers to keep it safe Data I will facilitate a new idea that you should send the work to the data, not the other way around. Data

WHY DO THIS? BECAUSE IT GETS THE ANSWERS SOOOO MUCH FASTER Name. Node Client

WHY DO THIS? BECAUSE IT GETS THE ANSWERS SOOOO MUCH FASTER Name. Node Client Copyright © 2013, SAS Institute Inc. All rights reserved.

WOW, that’s awesome. Can we join your cluster?

WOW, that’s awesome. Can we join your cluster?

We’ll be very good. Really.

We’ll be very good. Really.

Agenda 1. Basics 2. • • • Cluster Evolution Vanilla Cluster Foreign Workload Introduced

Agenda 1. Basics 2. • • • Cluster Evolution Vanilla Cluster Foreign Workload Introduced Node Specialization Cluster Specialization Datacenter Integration 3. YARN 4. Security

2012 : : Have Want

2012 : : Have Want

Vanilla Cluster Name. Node Sec. Nm. Node Data. Node

Vanilla Cluster Name. Node Sec. Nm. Node Data. Node

Vanilla Cluster (with foreign workload) Name. Node Sec. Nm. Node Data. Node

Vanilla Cluster (with foreign workload) Name. Node Sec. Nm. Node Data. Node

Foreign != Map. Reduce & not only ( SAS ) § SAS High Performance

Foreign != Map. Reduce & not only ( SAS ) § SAS High Performance Analytics § SAS Visual Analytics § Impala § BDAS Spark § Giraph § Solr §. . Hbase

Vanilla Cluster (with foreign workload) Name. Node Data. Node 1. 2. 3. 4. 5.

Vanilla Cluster (with foreign workload) Name. Node Data. Node 1. 2. 3. 4. 5. Add work across entire cluster Add memory to accommodate Sec. Nm. Node Derate Map. Reduce to accommodate Time Slice? Data. Node No extra copy of Data. Node

Node Specialization (for foreign workload) Name. Node Sec. Nm. Node Data. Node

Node Specialization (for foreign workload) Name. Node Sec. Nm. Node Data. Node

Node Specialization (for foreign workload) 1. 2. Name. Node 3. Data. Node 4. Data.

Node Specialization (for foreign workload) 1. 2. Name. Node 3. Data. Node 4. Data. Node 5. 6. Data. Node Add workload to some … “SASnodes” Add memory to SASnodes Sec. Nm. Node Derate Map. Reduce on SASnodes? Data. Node Cgroups to make em play nice Still no extra copy of Data. Node SAS writes data to SASnodes only. (balancer) Data. Node

Node Specialization (for foreign workload) 1. 2. Name. Node 3. Data. Node 4. Data.

Node Specialization (for foreign workload) 1. 2. Name. Node 3. Data. Node 4. Data. Node 5. 6. Data. Node Add workload to some … “SASnodes” Add memory to SASnodes Sec. Nm. Node Derate Map. Reduce on SASnodes? Data. Node Cgroups to make em play nice Still no extra copy of Data. Node SAS writes data to SASnodes only. (balancer) Data. Node CDH 4 Best Practice

Specialty Cluster Name. Node Sec. Nm. Node Data. Node Name. Node Data. Node

Specialty Cluster Name. Node Sec. Nm. Node Data. Node Name. Node Data. Node

Specialty Cluster 1. 2. Name. Node 3. Data. Node 4. Create new “Odd Shape”

Specialty Cluster 1. 2. Name. Node 3. Data. Node 4. Create new “Odd Shape” cluster Optimize Hardware to fit task Sec. Nm. Node Oops! extra copy of Data. Node Easier to contain variation Name. Node Data. Node

EXAMPLE ASYMMETRIC AS AN OPTION Name. Node Controller Client Copyright © 2013, SAS Institute

EXAMPLE ASYMMETRIC AS AN OPTION Name. Node Controller Client Copyright © 2013, SAS Institute Inc. All rights reserved.

Data. Center Integration CLIENT TERADATA HADOOP ORACLE

Data. Center Integration CLIENT TERADATA HADOOP ORACLE

Agenda 1. Basics 2. • • • Cluster Evolution Vanilla Cluster Foreign Workload Introduced

Agenda 1. Basics 2. • • • Cluster Evolution Vanilla Cluster Foreign Workload Introduced Node Specialization Cluster Specialization Datacenter Integration 3. YARN 4. Security

2013 q 4? 2014?

2013 q 4? 2014?

2013 q 4? 2014?

2013 q 4? 2014?

Node Specialization (for foreign workload) Name. Node Sec. Nm. Node Data. Node

Node Specialization (for foreign workload) Name. Node Sec. Nm. Node Data. Node

Agenda 1. Basics 2. • • • Cluster Evolution Vanilla Cluster Foreign Workload Introduced

Agenda 1. Basics 2. • • • Cluster Evolution Vanilla Cluster Foreign Workload Introduced Node Specialization Cluster Specialization Datacenter Integration 3. YARN 4. Security

Security is Hard. Better Start right away. § Add Kerberos to your environment ASAP

Security is Hard. Better Start right away. § Add Kerberos to your environment ASAP – right after the first POC § Integrate with the identity management on site - Don’t add unix-users to the cluster by hand! - Automate. - Engage SAS Technical Resources. - Security settings can be hard to get right. Error messages get obfuscated and tracking the true source is difficult - Easier to start with a small working system and add projects § Resist “Oh, we will add the security later”. Your users will have gotten so used to nosecurity they’l scream!

Thank You! Paul. Kent @ sas. com @hornpolish paulmkent

Thank You! Paul. Kent @ sas. com @hornpolish paulmkent