CompactPCI adds ruggedness and reliability to the standard PCI bus.
BY THOMAS SZALA
The explosion of data is straining equipment vendors' ability to keep pace and challenging them to deliver high-performance, high-availability systems. This is particularly true in the telecommunications market, which has seen a huge bandwidth explosion along with a proliferation of new service offerings. Proprietary approaches have become too expensive to develop and maintain, driving vendors and systems integrators to adopt open standards and mass-market technology to achieve time-to-market and cost economies.
For some applications, the solution for building reliable, high-performance storage systems requires a combination of industry standards such as the CompactPCI system bus architecture, RAID, and SCSI. These standards meet a broad range of price/performance goals.
Why CompactPCI RAID?
CompactPCI is a relatively new bus architecture that combines the features of several existing bus architectures to create a robust, reliable platform for embedded applications.
CompactPCI is based on the Peripheral Component Interconnect (PCI) standard but adds the rugged mechanical aspects of the IEC/IEEE Eurocard standards used in buses such as VMEbus. CompactPCI also makes use of the high-density IEC 2mm connector scheme popular in the telecommunications industry.
Figure 1: Service Control Points (SCPs) support the database storage functions in an SS7 telecommunications network.
The passive-backplane Compact PCI bus can provide a more reliable alternative to the motherboard-based PCI bus, and it's more versatile and inexpensive than VMEbus. In contrast to the edge-card connectors used in PCI, CompactPCI is based on a high-reliability pin-and-socket connector system with shielding for EMI/RFI protection. In contrast to VMEbus, with its limited-capacity DIN connector, CompactPCI uses a 2mm connector with a high pin count that allows I/O to be routed to the rear of chassis, which results in an uncluttered front panel and simplified servicing. It also provides a vehicle for auxiliary buses and user-defined functions.
For embedded applications where lost data and downtime are unacceptable, RAID has become the technique of choice. The use of RAID algorithms improves data accessibility and reliability.
Each RAID level uses one or more basic methodologies:
- Striping, which distributes data among the disks in an array;
- Mirroring, which stores identical data on multiple drives; and
- Parity, which calculates the data necessary to restore the contents on a drive if it fails and then stores that data on a different disk drive.
Each RAID level has trade-offs in performance, protection, and cost. System I/O throughput and fault-tolerant requirements determine which RAID level is appropriate.
One embedded CompactPCI/RAID controller application in telecommunications is the Service Control Point (SCP) function in an SS7 network. As shown in Figure 1, the SCP supports the database storage functions required in an SS7 network. These functions include subscriber billing and network maintenance information, intelligent network features (e.g., call waiting, voice mail, and caller ID), and special number translation services (e.g., 800/900 number processing). Typically, the SCP receives database storage-and-retrieval requests from the Signal Transfer Point (STP), processes those requests, and then either stores, or retrieves and forwards the appropriate information to the STP.
Due to their mission-critical nature, SCPs are deployed as redundant pairs, although not necessarily next to one another. Using the CompactPCI architecture, each SCP can be configured with a host CPU board, two WAN I/O boards, and an embedded RAID controller board-all of which are connected across a CompactPCI bus. For database storage-and-retrieval request servicing, SCPs communicate with STPs over WAN links. The RAID controller includes the Ultra2 SCSI channels that interface with the disk arrays. A disk array of up to 15 individually accessed disks can be attached to each Ultra2 SCSI channel on the RAID controller.
Figure 2: RAID 5 distributes parity data among the drives in a disk array.
SS7 database files are relatively small, and read/writes are frequent. RAID 5-disk striping with distributed parity-performs well with I/O request-intensive loads and is often the best RAID configuration for SCPs.
As shown in Figure 2, RAID 5 distributes parity data among the drives in a disk array, somewhat reducing the write bottleneck associated with dedicated parity drives. RAID 5 has excellent read performance, but its write performance suffers because it takes time to calculate and then write parity data. However, caching and other techniques can improve write performance. With RAID 5, if a drive fails, its contents can be reconstructed from the remaining drives.
CompactPCI-based RAID controllers can be an ideal solution for the telecommunications market. Database storage functions such as billing, voice messaging, and maintenance information, as well as data processing functions such as subscriber control and number translation, all benefit from the fault tolerance and I/O processing power of CompactPCI-based RAID controllers.
Thomas Szala is marketing director at Cyclone Microsystems (www.cyclone.com) in New Haven, CT.