Why are distributed systems hard?
Why are distributed systems hard?
Things fail, all the time!
Why are distributed systems hard?
They're BIG.
Why are distributed systems hard?
There's a lot of moving pieces.
Mesos is a
- Cluster Resource Manager
- Task Scheduler
Request model
|
2-level model
|
|
|
Mesos is...
| A top-level Apache project |
| A cluster resource negotiator |
| Scalable to 10,000s of nodes |
| Fault-tolerant, battle-tested |
| An SDK for distributed apps |
Why use Mesos for my next distributed system?
| Task distribution, launching, monitoring,
failure detection, killing, and cleanup |
| Resource isolation with containers |
| Messaging between tasks |
| Distributed state |
What is Mesos good for?
Fault tolerance
Fault Tolerance
Task Failure :(
Fault Tolerance
Local Agent Failure :(
Fault Tolerance
Host or Network Failure :(
Fault Tolerance
Mesos Master Failure :(
Fault Tolerance
Scheduler Failure :(
What is Mesos good for?
Long running services & batch jobs / analytics workloads
What is Mesos good for?
Elasticity
What is Mesos good for?
Drastically increasing utilisation
Connect to the master node
ssh jclouds@$master_ip
Download and inspect our convenience script
wget http://downloads.mesosphere.io/tutorials/spark_demo.sh
nano spark_demo.sh
Run it, saving results to a file
./spark_demo.sh > results.txt
As the job runs, Spark will print out details about what it’s doing
While the job is running, the Spark Web UI can be seen at http://$master_ip:4040
(note you must connect to the VPN)
While the job is running, the Spark Web UI can be seen at http://$master_ip:4040
(note you must connect to the VPN)
When the job is complete, the PageRank results can be viewed
nano results.txt
Thanks!
Come and talk to us
P.S., we're hiring!