Ceph@MIMOS: Growing Pains from R&D
to Deployment
Jing Yuan LUKE
Advanced Computing Lab
Ceph Day Kuala Lumpur, 22 August 2016
• MIMOS – A Brief Overview
• Distributed Object Storage in the era of Big Data
• Our Ceph Journey
• Moving Forward
• Concluding Remark
Copyright © 2016, MIMOS Bhd 2
Outline
3
MIMOS: An Overview
Enhancing ICT industry growth through
indigenous technologies
R&D in ICT
http://www.mimos.my
Copyright © 2016, MIMOS Bhd
MIMOS R&D Labs
Data Centric
Approach
4Copyright © 2016, MIMOS Bhd
MIMOS R&D Labs
Data Centric
Approach
5
INFORMATION
SECURITY
PHOTONICS
NANOELECTRONICS
ADVANCED
ANALYSIS &
MODELING
ARTIFICIAL
INTELLIGENCE
WIRELESS
COMMUNICATIONS
ADVANCED
INFORMATICS
ADVANCED
COMPUTING
MICRO-ELECTRONICS
/ENERGY
ACCELERATIVE
TECHNOLOGY
INTELLIGENT
INFORMATICS
USER
EXPERIENCE
Copyright © 2016, MIMOS Bhd
Copyright © 2016, MIMOS Bhd 6
Enterprise
DataStore
Structured Data
DATA Sources in the Big Data Era
Enterprise
DataStore
Structured Data
Copyright © 2016, MIMOS Bhd 7
Sensors
Social
Networks
Unstructured
Data
Mobilie
Devices
Wearables
Tags
DATA Sources in the Big Data Era
DATA
The new Currency
Enterprise
DataStore
Structured Data
Copyright © 2016, MIMOS Bhd 8
Sensors
Social
Networks
Unstructured
Data
Mobilie
Devices
Wearables
The TSUNAMI
Tags
Copyright © 2016, MIMOS Bhd
9
DATA
The new Currency
Knowledge
The new Discovery
Enlightenment
Copyright © 2016, MIMOS Bhd
10
DATA
The new Currency
Knowledge
The new Discovery
Enlightenment
More data generated
More storage required
Copyright © 2016, MIMOS Bhd 11
Distributed Object Storage in Big Data Era
Source: SNIA : Swift Object Storage adding EC (Erasure Code), 2014
Copyright © 2016, MIMOS Bhd 12
Distributed Object Storage in Big Data Era
Source: SNIA : Swift Object Storage adding EC (Erasure Code), 2014
• Other potential benefits
• Scales
• Capability
• Capacity
• Hardware agnostics
• Self-healing
• Replication
• Erasure Codes
Copyright © 2016, MIMOS Bhd 13
Our Ceph Journey
2013
2014
2015
2016
• R&D
• PoC/Testbed
Copyright © 2016, MIMOS Bhd 14
Our Ceph Journey
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Which solution/platform that can provide:
– Support to existing cloud initiatives, is it well
accepted by:
• OpenStack
• OpenNebula
• Others
• Provide different ways to access the backend
– Web services
– Block like access
– File System
• Highly Available
– Active-active
• Linux upstream support
Copyright © 2016, MIMOS Bhd 15
Our Ceph Journey
2013
2014
2015
2016
• R&D
• PoC/Testbed
12
3 MIMOS
TPMHPCC2 HPCC1
321
MIMOS
KHTP
Copyright © 2016, MIMOS Bhd 16
Our Ceph Journey
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Big Data Storage
Event
• First Internal
Deployment
Copyright © 2016, MIMOS Bhd 17
Our Ceph Journey
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Big Data Storage
Event
• First Internal
Deployment
Big Data Storage @ Big Data Week KL 2014
Copyright © 2016, MIMOS Bhd 18
Our Ceph Journey
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Big Data Storage
Event
• First Internal
Deployment
• Simple Backup/Archiving
– First attempt to use Ceph in a small
production cluster
– Backup application: BackupPC
– Access: CephFS
– Challenge: BackupPC creates a lot of small
files (several kB), saw plenty of space wasted
(due to the default 4MB object size), solution
create a dedicated pool and use extended
attributes to assign mount point to different
pool and reduce object size
– Moving forward: considering RBD and Bacula
Copyright © 2016, MIMOS Bhd 19
Our Ceph Journey
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Big Data Storage
Event
• First Internal
Deployment
• Mi-ROSS 1.0
development
• More
deployments for:
– VDI
– Government
agencies, law
enforcement
agency, etc.
– Expanding
internal cluster
Copyright © 2016, MIMOS Bhd 20
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Big Data Storage
Event
• First Internal
Deployment
• Mi-ROSS 1.0
development
• More
deployments for:
– VDI
– Government
agencies, law
enforcement
agency, etc.
– Expanding
internal cluster
Our Ceph Journey
Copyright © 2016, MIMOS Bhd 21
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Big Data Storage
Event
• First Internal
Deployment
• Mi-ROSS 1.0
development
• More
deployments for:
– VDI
– Government
agencies, law
enforcement
agency, etc.
– Expanding
internal cluster
• VDI
– Challenges:
• 60+ Windows based VMs accessed via
RDP, for s/w development environment,
lots of I/O (code check-in/out, compiling,
etc.)
• 30+ development VMs
– Solution: Used a experimental feature, i.e. KV
based datastore instead of the typical journal
based datastore
Our Ceph Journey
Copyright © 2016, MIMOS Bhd 22
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Big Data Storage
Event
• First Internal
Deployment
• Mi-ROSS 1.0
development
• More
deployments for:
– VDI
– Government
agencies, law
enforcement
agency, etc.
– Expanding
internal cluster
• Storage for Cloud Deployment and General Uses
– Mi-Cloud (MIMOS Cloud Platform)
• Ceph provides datastores for and to VMs
– Support for multiple workloads via Mi-ROSS
• File Sharing (NFS and SAMBA)
MyHDW
Our Ceph Journey
Copyright © 2016, MIMOS Bhd 23
Our Ceph Journey
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Big Data Storage
Event
• First Internal
Deployment
• Mi-ROSS 1.0
development
• More
deployments for:
– VDI
– Government
agencies, law
enforcement
agency, etc.
– Expanding
internal cluster
• Mi-ROSS 2.0
development
• Both Mi-Cloud
and Mi-ROSS
supporting 600+
VMs internally
Copyright © 2016, MIMOS Bhd
2013
2014
2015
2016
• R&D
• PoC/Testbed
• Big Data Storage
Event
• First Internal
Deployment
• Mi-ROSS 1.0
development
• More
deployments for:
– VDI
– Government
agencies, law
enforcement
agency, etc.
– Expanding
internal cluster
• Mi-ROSS 2.0
development
• Both Mi-Cloud
and Mi-ROSS
supporting 600+
VMs internally
24
Our Ceph Journey
• As we grow we have learnt about various operation issues, e.g.
– Impact on performance as we add/remove disks into the cluster
• How to work around it
– Dealing with different network switches
• Different brands/models often implement the same protocols slight differently
– Monitoring and Maintenance
• Developing monitoring agents/plug-ins
• Developing our own SOP
– Fine tuning
• Ceph
• Kernel
Copyright © 2016, MIMOS Bhd 25
Lessons learnt from the Journey
• A NAS appliance like building on top of Ceph providing:
– Ceph Management
• Pool
• RBD
• CRUSH (planned)
• Keys (planned)
– File sharing
• SAMBA
• NFS
• iSCSI (in progress)
– Dashboard
• Utilization
• Health
• Other Ceph related statistics
– Other Storage Services
• “Dropbox” (planned)
• Backup (planned)
Copyright © 2016, MIMOS Bhd 26
Mi-ROSS
• Distributed Storage for Disaster
Mitigation and Smart Cities
• How can distributed block
storage such Ceph and be used
to address Edge Computing
– Network Latency
– Ad-Hoc-ness of the edge devices
– Etc
Copyright © 2016, MIMOS Bhd 27
Moving Forward
• Distributed Object Storage such Ceph can address the data deluge
from the era of Big Data and IoT
• MIMOS’ Advance Computing Lab would like welcome all to collaborate
to further enhance this exciting open solutions
Copyright © 2016, MIMOS Bhd 28
Concluding Remark
Copyright © 2016, MIMOS Bhd 29
Thank You

Ceph@MIMOS: Growing Pains from R&D to Deployment

  • 1.
    Ceph@MIMOS: Growing Painsfrom R&D to Deployment Jing Yuan LUKE Advanced Computing Lab Ceph Day Kuala Lumpur, 22 August 2016
  • 2.
    • MIMOS –A Brief Overview • Distributed Object Storage in the era of Big Data • Our Ceph Journey • Moving Forward • Concluding Remark Copyright © 2016, MIMOS Bhd 2 Outline
  • 3.
    3 MIMOS: An Overview EnhancingICT industry growth through indigenous technologies R&D in ICT http://www.mimos.my Copyright © 2016, MIMOS Bhd
  • 4.
    MIMOS R&D Labs DataCentric Approach 4Copyright © 2016, MIMOS Bhd
  • 5.
    MIMOS R&D Labs DataCentric Approach 5 INFORMATION SECURITY PHOTONICS NANOELECTRONICS ADVANCED ANALYSIS & MODELING ARTIFICIAL INTELLIGENCE WIRELESS COMMUNICATIONS ADVANCED INFORMATICS ADVANCED COMPUTING MICRO-ELECTRONICS /ENERGY ACCELERATIVE TECHNOLOGY INTELLIGENT INFORMATICS USER EXPERIENCE Copyright © 2016, MIMOS Bhd
  • 6.
    Copyright © 2016,MIMOS Bhd 6 Enterprise DataStore Structured Data DATA Sources in the Big Data Era
  • 7.
    Enterprise DataStore Structured Data Copyright ©2016, MIMOS Bhd 7 Sensors Social Networks Unstructured Data Mobilie Devices Wearables Tags DATA Sources in the Big Data Era
  • 8.
    DATA The new Currency Enterprise DataStore StructuredData Copyright © 2016, MIMOS Bhd 8 Sensors Social Networks Unstructured Data Mobilie Devices Wearables The TSUNAMI Tags
  • 9.
    Copyright © 2016,MIMOS Bhd 9 DATA The new Currency Knowledge The new Discovery Enlightenment
  • 10.
    Copyright © 2016,MIMOS Bhd 10 DATA The new Currency Knowledge The new Discovery Enlightenment More data generated More storage required
  • 11.
    Copyright © 2016,MIMOS Bhd 11 Distributed Object Storage in Big Data Era Source: SNIA : Swift Object Storage adding EC (Erasure Code), 2014
  • 12.
    Copyright © 2016,MIMOS Bhd 12 Distributed Object Storage in Big Data Era Source: SNIA : Swift Object Storage adding EC (Erasure Code), 2014 • Other potential benefits • Scales • Capability • Capacity • Hardware agnostics • Self-healing • Replication • Erasure Codes
  • 13.
    Copyright © 2016,MIMOS Bhd 13 Our Ceph Journey 2013 2014 2015 2016 • R&D • PoC/Testbed
  • 14.
    Copyright © 2016,MIMOS Bhd 14 Our Ceph Journey 2013 2014 2015 2016 • R&D • PoC/Testbed • Which solution/platform that can provide: – Support to existing cloud initiatives, is it well accepted by: • OpenStack • OpenNebula • Others • Provide different ways to access the backend – Web services – Block like access – File System • Highly Available – Active-active • Linux upstream support
  • 15.
    Copyright © 2016,MIMOS Bhd 15 Our Ceph Journey 2013 2014 2015 2016 • R&D • PoC/Testbed 12 3 MIMOS TPMHPCC2 HPCC1 321 MIMOS KHTP
  • 16.
    Copyright © 2016,MIMOS Bhd 16 Our Ceph Journey 2013 2014 2015 2016 • R&D • PoC/Testbed • Big Data Storage Event • First Internal Deployment
  • 17.
    Copyright © 2016,MIMOS Bhd 17 Our Ceph Journey 2013 2014 2015 2016 • R&D • PoC/Testbed • Big Data Storage Event • First Internal Deployment Big Data Storage @ Big Data Week KL 2014
  • 18.
    Copyright © 2016,MIMOS Bhd 18 Our Ceph Journey 2013 2014 2015 2016 • R&D • PoC/Testbed • Big Data Storage Event • First Internal Deployment • Simple Backup/Archiving – First attempt to use Ceph in a small production cluster – Backup application: BackupPC – Access: CephFS – Challenge: BackupPC creates a lot of small files (several kB), saw plenty of space wasted (due to the default 4MB object size), solution create a dedicated pool and use extended attributes to assign mount point to different pool and reduce object size – Moving forward: considering RBD and Bacula
  • 19.
    Copyright © 2016,MIMOS Bhd 19 Our Ceph Journey 2013 2014 2015 2016 • R&D • PoC/Testbed • Big Data Storage Event • First Internal Deployment • Mi-ROSS 1.0 development • More deployments for: – VDI – Government agencies, law enforcement agency, etc. – Expanding internal cluster
  • 20.
    Copyright © 2016,MIMOS Bhd 20 2013 2014 2015 2016 • R&D • PoC/Testbed • Big Data Storage Event • First Internal Deployment • Mi-ROSS 1.0 development • More deployments for: – VDI – Government agencies, law enforcement agency, etc. – Expanding internal cluster Our Ceph Journey
  • 21.
    Copyright © 2016,MIMOS Bhd 21 2013 2014 2015 2016 • R&D • PoC/Testbed • Big Data Storage Event • First Internal Deployment • Mi-ROSS 1.0 development • More deployments for: – VDI – Government agencies, law enforcement agency, etc. – Expanding internal cluster • VDI – Challenges: • 60+ Windows based VMs accessed via RDP, for s/w development environment, lots of I/O (code check-in/out, compiling, etc.) • 30+ development VMs – Solution: Used a experimental feature, i.e. KV based datastore instead of the typical journal based datastore Our Ceph Journey
  • 22.
    Copyright © 2016,MIMOS Bhd 22 2013 2014 2015 2016 • R&D • PoC/Testbed • Big Data Storage Event • First Internal Deployment • Mi-ROSS 1.0 development • More deployments for: – VDI – Government agencies, law enforcement agency, etc. – Expanding internal cluster • Storage for Cloud Deployment and General Uses – Mi-Cloud (MIMOS Cloud Platform) • Ceph provides datastores for and to VMs – Support for multiple workloads via Mi-ROSS • File Sharing (NFS and SAMBA) MyHDW Our Ceph Journey
  • 23.
    Copyright © 2016,MIMOS Bhd 23 Our Ceph Journey 2013 2014 2015 2016 • R&D • PoC/Testbed • Big Data Storage Event • First Internal Deployment • Mi-ROSS 1.0 development • More deployments for: – VDI – Government agencies, law enforcement agency, etc. – Expanding internal cluster • Mi-ROSS 2.0 development • Both Mi-Cloud and Mi-ROSS supporting 600+ VMs internally
  • 24.
    Copyright © 2016,MIMOS Bhd 2013 2014 2015 2016 • R&D • PoC/Testbed • Big Data Storage Event • First Internal Deployment • Mi-ROSS 1.0 development • More deployments for: – VDI – Government agencies, law enforcement agency, etc. – Expanding internal cluster • Mi-ROSS 2.0 development • Both Mi-Cloud and Mi-ROSS supporting 600+ VMs internally 24 Our Ceph Journey
  • 25.
    • As wegrow we have learnt about various operation issues, e.g. – Impact on performance as we add/remove disks into the cluster • How to work around it – Dealing with different network switches • Different brands/models often implement the same protocols slight differently – Monitoring and Maintenance • Developing monitoring agents/plug-ins • Developing our own SOP – Fine tuning • Ceph • Kernel Copyright © 2016, MIMOS Bhd 25 Lessons learnt from the Journey
  • 26.
    • A NASappliance like building on top of Ceph providing: – Ceph Management • Pool • RBD • CRUSH (planned) • Keys (planned) – File sharing • SAMBA • NFS • iSCSI (in progress) – Dashboard • Utilization • Health • Other Ceph related statistics – Other Storage Services • “Dropbox” (planned) • Backup (planned) Copyright © 2016, MIMOS Bhd 26 Mi-ROSS
  • 27.
    • Distributed Storagefor Disaster Mitigation and Smart Cities • How can distributed block storage such Ceph and be used to address Edge Computing – Network Latency – Ad-Hoc-ness of the edge devices – Etc Copyright © 2016, MIMOS Bhd 27 Moving Forward
  • 28.
    • Distributed ObjectStorage such Ceph can address the data deluge from the era of Big Data and IoT • MIMOS’ Advance Computing Lab would like welcome all to collaborate to further enhance this exciting open solutions Copyright © 2016, MIMOS Bhd 28 Concluding Remark
  • 29.
    Copyright © 2016,MIMOS Bhd 29 Thank You