Our Disaster Recovery (DR) site is hosted at a carrier-neutral datacenter in Uppsala managed by Klövern. It is located far from our primary site to make sure that backups and standby hosts will not be affected in the unlikely event of a major problem with the primary site. Some services with exceptionally high availability requirements, such as DNS- and Authentication-servers, are kept running at both sites continuously.
Both locations feature secure sites with 24×7 manned security presence and perimeter intruder monitoring.
The two sites are connected through two discrete gigabit connections and the entire network in each site is built with fully redundant connections to all servers and equipment, and with no single points of failure.
Internet connections (Transit & Peering)
Our network is also connected to STHIX (Stockholm Internet Exchange) for public peering. Some of our larger peering partners are Hurricane Electric, Elisa Oyj, Comhem, DNA Oy, Alltele, and Stockholm Stadsnät.
We are using Noction Intelligent Routing Platform to automatically monitor our network connections and route the traffic using the best performing path and to avoid network trouble. This guarantees that traffic from our servers always will take the best route provided by our transit partners and peering partners.
Reference: AS-RACKFISH , AS49232 at ripedb.
For high performance storage we use Nimble Storage 10GBE storage cabinets with flash based caching of heavily used content. We also deploy Dell EqualLogic 1GBE for internal use, backups and low-cost high volume storage and older legacy servers with multi-site data replication.
Every storage shelf has two controllers (Active/Standby) so that we easily may upgrade firmware and perform controlled failover tests without loosing connectivity to our storage. Every physical VM host is connected with separate network adapters, through separate network switches to separate network adapters in the storage controllers.
RAID is configured with double parity (RAID6) and with a hot spares.
Depending on customer needs we do point in time snapshots locally and replication of storage to our DR site.
We do failover tests between storage controllers once a month.
Daily backups are stored for 7 days, weekly for 5 weeks and monthly backups are stored for one year. I addition to this, most of our services also have multiple daily snapshots.
Restores are performed using standard protocols such as iSCSI, and randomly selected data recovery tests are performed monthly while full virtual machine recovery tests are performed quarterly.