01.02.2013 Views

Chapter 4 PA-RISC Computer Systems - OpenPA.net

Chapter 4 PA-RISC Computer Systems - OpenPA.net

Chapter 4 PA-RISC Computer Systems - OpenPA.net

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Stratus Continuum Hardware Details<br />

(ISDN, serial, parallel, X.25, and all sorts of other comm boards). HP-UX did not support many of<br />

those, if any. VOS also supported disk and tape I/O through this.<br />

Expansion chassis is available to house additional secondary I/O cards or disk shelves (for large disk<br />

farms, etc).<br />

Architecture<br />

Each logical processor is physically two pairs of actual CPUs (that means four physical CPU chips per<br />

single logical one).<br />

Each pair is located on a separate FRU. All processors run “lock-stepped” (that is, they do exactly<br />

the same thing at the same time). Comparator logic between each two physical CPU pair monitors<br />

for discrepancies. If any physical CPU glitches or does something different, the comparator logic will<br />

detect the error and take that pair of CPUs offline, while the system continues to run on the other pair.<br />

There is no “failover time.” On multi-processor boards, each FRU contains multiple pairs of the logical<br />

processor halves.<br />

The memory is self-checking and ECC corrected. If an uncorrectable error occurs, the FRU in which<br />

the memory is located will also be taken offline.<br />

The DMA engines for the big I/O boards is designed such that the main memory content is protected<br />

from an errant card from scribbling over addresses that it was not supposed to write to. This of course<br />

is programmed by the OS device drivers.<br />

The big I/O boards are also self-checking and contain a pair of everything. However, with the exception<br />

of the K600 they do not run lock-stepped to the twin FRU. For example on the K450/K460 boards,<br />

each of the SCSI host adapters is connected via the backplane into the same SCSI bus on the partner<br />

board, but each board’s controller occupies a different SCSI target ID. Only one controller is normally<br />

active, but when a failure occurs on the active board, all I/O is switched to the other controller. For the<br />

Ether<strong>net</strong> ports on that board, they can be wired up to the same <strong>net</strong>work or to different <strong>net</strong>works, and<br />

a software RNI (redundant <strong>net</strong>work interface) layer provides transparent switching.<br />

Other communications interfaces employ software-driven failover schemes.<br />

All disks are mirrored. Early FTX 3.x releases used an in-house virtual disk layer (VDL) driver, but<br />

later releases switched to a modified version of the Veritas VxVM product. In HP-UX, HP’s own LVM<br />

(logical volume manager) is used. VOS, of course, has its own disk mirroring scheme.<br />

The Continuum 400 series has the same CPU/memory architecture as the 600/1200, but the I/O bus<br />

is different. Instead of a Golf bus, it has an X bus that connects each CPU/memory module to a pair<br />

of PCI bridge boards. All I/O connectivity is via PCI cards. There are two PCI bays of 7 slots each,<br />

connected downstream from the PCI bridge boards. Each bay has a dual channel SCSI adapter on it as<br />

standard equipment. These are also cross-wired and dual-initiated much in the same way as the SCSI<br />

ports on the 600/1200 systems. The 400 is also typically shipped with a pair of Ether<strong>net</strong> adapter cards.<br />

The PCI bridge boards also each contains a removable PCMCIA flash memory card. This is used as the<br />

boot device. FTX puts the bootloader as well as the UNIX kernel on there, whereas HP-UX only uses<br />

it for the bootloader.<br />

The PCI bay doors control the power the the PCI slots. Once opened, all slots in that bay are powered<br />

off to facilitate removal and insertion of cards. The system continues to run on cards in the other bay.<br />

An interlock mechanism prevents both bay doors from being opened at the same time.<br />

Again, all disks are mirrored as they are on the 600/1200 series, and communications interfaces use<br />

software-controlled failover mechanisms.<br />

351

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!