CU Home
Columbia University Information Technology
Backup Service Hardware

Software > Connected DataProtector > Backup Service Hardware

Production Server Configuration

Three servers (one for standby) configured as follows:
  • two HP ProLiant DL380 G3's
  • two Intel Xeon 2.8 GHz processors
  • 512KB cache
  • 4GB Ram
  • two 36.4GB and two 146.8GB disks
  • one scsi attached Promise VTrak 15100 15-drive Raid 5 array with 3.25 TB (max per box). Each server can connect a maximum of six 15100's which would service approximately 9750 users in a fully expanded configuration.
  • one Plextor 708A DVD burner
  • 3-year onsite warranty
The Connected Dataprotector hardware infrastructure consists of two identical server systems co-located in the main machine room and at 103 Philosophy. Each system contains identical copies of the backed up user data as well as system and application software. The overall design of the service allows either system (but not both) to be down without any interruption of service to users. When a system that has been down is restored to service, any data that may have been sent to the other server is replicated on the newly enabled system.

Each system consists of the following: an HP DL380 G3 dual processor server running Windows Server 2003 with a scsi attached Promise Vtrak 11500 external raid box. The internal server disks are mirrored so that all control and program information is duplicated on separate drives.

The Promise Vtrak holds a maximum of 3.75TB of data on 15 Hitachi 250GB SATA disk drives. The box has dual power supplies and cooling fans, and all disks are hot swappable. Of these 15 drives, one is allocated as a hot spare (in case a disk fails) and two are used to hold raid parity information. A raid 50 scheme is used which divides the 14 available drives into two groups of seven, each group having its own parity drive (much like two raid 5 groups). Data that is backed up is striped across the two groups. If a disk fails in either group, the parity information on the parity disk in that group allows the system to reconstruct the missing data in real-time so that applications are unaware of the failure (response time may be slower). After a disk failure has occurred the Vtrak rebuilds the missing data on the hot spare (this may take 3 to 4 hours depending on the amount of data on the disk), thus restoring the capability of the raid group to tolerate another disk failure. Each raid group can tolerate a single disk failure until the failed disk is rebuilt on the hot spare; after the rebuild, it can tolerate another failure. If an additional failure occurs in a raid group (highly unlikely) before the rebuild completes, then there will be data loss. However, as this backup service duplicates the raid function in two physically separate locations, the likelihood of data loss is greatly diminished.

« Back to Connected Homepage