Difference between revisions of "User:Chris Tyler/Fedora ARM Infra Redesign"

From CDOT Wiki
Jump to: navigation, search
(Current Arrangement)
(Recommended Reallocation)
Line 77: Line 77:
 
** Move koji-web to a separate host (Japan? OpenRD?)
 
** Move koji-web to a separate host (Japan? OpenRD?)
 
** Add additional createrepo builders (Japan? NewZealand?)
 
** Add additional createrepo builders (Japan? NewZealand?)
** Move NFS activity to guruplugs
+
** Move NFS activity to guruplugs (they're weak builders, but have good disk I/O (eSATA), could use multiple NFS servers with each supplying scratch space to one group of v7 builders)
 
 
  
 
== Scenarios for Quotation ==
 
== Scenarios for Quotation ==

Revision as of 12:08, 2 December 2011

Important.png
This is a draft only!
It is still under construction and content may change. Do not rely on this information.

Design Meeting 2011-12-01

It is apparent that the Fedora ARM build farm infrastructure needs to be upgraded to increase storage throughput, IOPS, and network utilization.

Current Arrangement

  • HongKong
    • Hardware:
      • Intel Core i7 Quad
      • 4 x 2TB disk
      • 2 x 128GB SATA SSD
      • 12 GB RAM
    • Hosting:
      1. Koji Web
      2. Koji Hub
      3. NFS export of /mnt/koji
      4. Kojid for createrepo channel
      5. DHCP & NAT for builders


  • Australia
    • Hardware:
      • Intel Q660 Quad
      • 2 x 1TB disk
      • 8 GB RAM
    • Hosting:
      1. NFS export for builder scratch space


  • Ireland
    • Hardware:
      • Intel Q6600 Quad (?)
      • 8 GB RAM
      • 2 x 2TB disk
    • Hosting:
      1. Sigul signing server (in VMs)

Recommended Reallocation

  • HongKong
    • Storage:
      • 4 x 2 TB in RAID 0+1
    • Hosting:
      1. Koji Web
      2. Koji Hub


  • Australia
    • Storage:
      • 4 x 2TB in RAID 0+1
      • Consider expansion to 8 drives
    • Hosting:
      1. NFS export for builder scratch space


  • Ireland
    • Storage:
      • 2 x 128GB SATA SSD
      • 2 x 2TB disk in RAID-1
    • Hosting:
      1. Postgresql DB


  • Chile (move from Raspi cubicle)
    • Storage:
      • 4 x 2TB disk in RAID 0+1
      • OR 2 x 2TB disk in RAID 0+1, plus 1 RevoDrive 3 x2 [MaxIOPS] 960G, with plans to add another RevoDrive in 2012, plus possibly some of the 1TB drives taken from Australia for local backup of the SSD (possibly RAID 1 with writemostly)
    • Hosting:
      1. NFS export of /mnt/koji
      2. kojid for createrepo


  • Possible Future Directions
    • Move koji-web to a separate host (Japan? OpenRD?)
    • Add additional createrepo builders (Japan? NewZealand?)
    • Move NFS activity to guruplugs (they're weak builders, but have good disk I/O (eSATA), could use multiple NFS servers with each supplying scratch space to one group of v7 builders)

Scenarios for Quotation

Bold indicates most likely scenarios


  • Scenario 1
    • 6 x 2TB disk (4 for Australia, 2 for Chile)
    • 2 x NIC


  • Scenario 2
    • 16 x 2TB disk (8 Australia, 6 Chile, 2 spares)
    • Multi-port SATA cards for Australia, Chile
    • 2 x NIC


  • Scenario 2.5
    • 6 x 2TB disk (4 Austrlia, 2 Chile)
    • 1 x RevoDisk 3 x2 960GB (Chile)
    • 2 x NIC


  • Scenario 3
    • 10 x 2 TB disk (8 Australia, 2 Chile)
    • Multi-port SATA card for Australia
    • 2 x RevoDisk 3 x2 960GB (Chile)
    • 2 x NIC


  • All cases add:
    • 5 Workstation NICs
    • Serial USB adapters