PCI-HIPPI Interfaces in the PMC form-factor.
Arie Van Praag, CERN EDU-DQ
Alberto Guglielmi, DIGITAL
Development Status
A first series of prototypes following exactly the PMC specifications was made and ready in March 1997. The schematics diagrams of the PCI-HIPPI Interfaces ware followed as much as possible. Changes that ware necessary consisted in a different type of FIFO and in the power-converters. To fit in the available space this converters had to be developed from scratch. The HIPPI connector had to be mounted bare without its metal mounting bracket. According to the HIPPI specification this may give a problem as the bracket is used for ground connection. However in practise we did not encounter any of form of ground problems.
Data Rates
Measurements showed that the PMC modules worked in every sense the same as the larger PCI boards, with the following data rates for ASCII data.
Host |
|
Source |
Destination |
Alpha 400 (Avanti) |
via converter board |
25 MBytes/s |
72 MBytes/s |
Alpha 500 (Maverick) |
via converter board |
25 MBytes/s |
91 MBytes/s |
Digital AXBVME 160 |
PMC |
24 Mbytes/s |
65 Mbytes/s |
Motorola VME Power PC |
PMC |
21 Mbytes/s |
58 Mbytes/s |
Problems encountered
The main problem not solved for the moment is mechanical.. The standard HIPPI-Connector is to big to be mounted on a PMC frontpanel.. Using no frontpanel makes mounting the interface in a VME unit unstable. Two solutions ware discussed:
- A smaller connector with an intermediate part to the cable connector
-
A extended printed circuit that brings the connector out in front
The last one was accepted if more modules are needed and a industrial production would have place.
Due to the PMC standards space restrictions use of cooling fins on the HIPPI Circuits is excluded. For the Destination this never was a problem. For the Source it shows up that the forced cooling in VME crates is so much better as in a desktop computer that we can do without special measures.
Future and Productions
The modules are made available to the CERN experiment groups where specially ATLAS and CMS are interested.
Contacts with Industry are made and INCAA Computers in the Netherlands are willing to take up a production as soon as sufficient interest is shown by the future CERN experiments or by other VME users.
This is one of the CERN High Speed Interconnect pages - 23 Septembre 1997 - Arie van Praag