Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Scale/SWITCHengines—IaaS for the Swiss Academic Community

Talk from a meetup of the Swiss OpenStack User Group

  • Login to see the comments

Scale/SWITCHengines—IaaS for the Swiss Academic Community

  1. 1. SCALE/SWITCHengines IaaS for the Swiss Academic Community CHOSUG meeting, Zurich, 5 March 2015 Simon Leinen
  2. 2. © 2015 SWITCH • SWITCH: National Research & Education Network (NREN) –non-profit foundation to provide ICT to academic community, est.1987 –largely funded by service fees, “upward delegation” from universities –10/100Gb/s IPv4/v6 backbone, SAML-based national AAI, many other services (lecture recording, spam marking etc.) • Cloud Project –Goal: Build modern infrastructure for “community cloud” offerings: • SaaS: “SWITCHdrive”–ownCloud-based sync & share • IaaS: “SWITCHengines”–OpenStack/Ceph-based VM/storage on demand • … –National co-funding from CUS P-2 program as project “SCALE” Background 2
  3. 3. © 2015 SWITCH • Two locations (University DCs Lausanne/Zurich), each with –32 2RU dual-Xeon (E5-2650v2) + 128 GB RAM + 2*10GE + 2*SSD –16 servers also have 12*4TB 3.5” disks  Ceph OSDs –2*48-port 10GE (+6-port 40GE) switches + 1*48-port GigE –Uplink: 2*10GE w/BGP-4 (IPv4+IPv6) directly to backbone • Currently two racks used per location –Each can grow up to ~20 racks • Plus a staging setup with two (tiny) sites –in one of the two production locations The Infrastructure 3
  4. 4. © 2015 SWITCH • Icehouse –Installed using Foreman + Puppet (puppet-openstack) –Keystone, Nova, Glance, Cinder, Neutron, Heat, Ceilometer(?) –RBD (Ceph) for all block storage • Using RBD integration patch • Controller functions in “pet” service VMs on “prov” hosts –prov hosts = two boxes per site in DRBD/LVM setup –service VMs: controller, network, foreman, dhcp, … • Dual-site setup –IPv6: all routable / IPv4: mostly RFC 1918, API/ext nets routable –Separate Ceph clusters –Separate regions (no cells) OpenStack Installation 4
  5. 5. © 2015 SWITCH • Everybody who uses L3 overlay (GRE/…) for tenant isolation seems to run into this • Standard “fix” seems to be to decrease tenant MTU –I don’t believe in this – 1500 bytes is too ingrained in the Internet now • We increased the underlay MTU (to 1600 bytes) • Everything starts to work nicely  highly recommended The MTU Issue 5
  6. 6. © 2015 SWITCH • Simpler Dashboard for simple uses –Working with external supplier • IPv6 support (should be easy with Juno!?) • Scale-out network (leaf/spine) beyond two racks • Push performance boundaries of OVS+VXLAN –Keeping the many alternatives in mind – e.g. Snabb Switch • “VPC” – VPN between VMs and customer campus Future Plans (Infrastructure) 6
  7. 7. © 2015 SWITCH • Elaborate suitable business/charging model(s) –with both university IT services and researchers • Proposed follow-up project “SCALE-UP” –Develop higher-level services in collaboration with universities: • Distributed computing/big data • Container management • Virtual Private Cloud • “Academic” SaaS • … Future Plans for SCALE/SWITCHengines 7
  8. 8. © 2015 SWITCH • We’re hiring! –see Questions? 8