1. Bob Hirosky L2  eta Review 26-APR-01 L2  eta Introduction L2  etas – a stable source of processing power for DØ Level2 Goals: Commercial (replaceable) components Provide replacement/upgrade path Match requirements for L2Alphas Minimize impact to experiment’s manpower hardware development online software commissioning effort Fastest possible development time (timescale of RunIIa) http://galileo.phys.virginia.edu/~rjh2j/l2beta

2. Bob Hirosky L2  eta Review 26-APR-01 Design Approach Hard drive 6U board 9U board mezzanine board FPGA Break off CPU functions into commercial SBC Implement custom I/O to as large an extent as possible in Firmware Use factory-made PCI interface, to lighten firmware complexity

3. Bob Hirosky L2  eta Review 26-APR-01 Functional Requirements Our aim is to develop a system that is both a functional match for the L2Alphas and supports both hardware and software compatibility to minimize impact on the trigger commissioning efforts. Functional requirements can be broken down into these categories: Processing power OS/Software tools VME Support Custom I/O (TSI, MBus) Mechanical and Electrical

4. Bob Hirosky L2  eta Review 26-APR-01 CPU Requirements Processing power 500 MHz 21164 2 integer, 2 floating pipes; 64 bits wide (used in bit searches and PIO) 96kB on-chip L2 cache; 3-way associative enough cache for data for event + enough instructions for 24 μsec of non-looping execution (L3 cache (4MB) does not work, doesn’t SEEM to be needed because of characteristics of L2 cache above) SI9515SF95 21 Executes L2 algorithms < 50μsec/event

5. Bob Hirosky L2  eta Review 26-APR-01 CPUSpecint95Specfp95 Alpha 500 MHz1521 PIII 800 MHz3829 PIII 850 MHz4135 PIII 933 MHz4539 PIII 1000MHz4841 Performance Benchmarks for common CPUs Integer performance is key for most L2 processing Tests include: running of compilers/interpreters, compression algorithms, data base, & chip simulations

6. Bob Hirosky L2  eta Review 26-APR-01 Modern CPUs performs well in comparison PIII on chip Cache = 256K vs 96K in Alpha, room for appx. 3 times the instructions with an event in memory. Faster memory access w/ modern RAM+controllers PIII SBCs readily available – CPU section debugging is a warranty issue!

7. Bob Hirosky L2  eta Review 26-APR-01 SBC of Choice VMIC 7740(50) PIII – 850 MHZ 100MHz x 64 bits RAM Bus Integrated UII ( VME support built in!) Single Slot Design PMC expansion site Single Vendor? NO! (Concurrent Tech + SBS/Bit3 * )

8. Bob Hirosky L2  eta Review 26-APR-01 Advantages of PIII – based CPU Popular commercial products = support Ongoing product development Mature LINUX/OS support Commercial, well supported, software readily available KAI C++ Match to Alpha byte ordering Closeness to Alpha Linux code Software development minimized (short list to SW compatibility) - rewrite device drivers - install memory access module (Alpha version was developed from Intel Linux) maybe, but maybe not necessary - interrupt handling : requires some thought, but Alpha experience should provide a big jumpstart

9. Bob Hirosky L2  eta Review 26-APR-01 Custom I/O Functions Strongest functional requirement is DMA: ~80-100 MB/s Data rate from MBus (FIFOs) to CPU Memory required Also short latency times for MBus programmed I/O even during DMA transfers Ability to generate PCI interrupts Provide register sets/functions compatible with L2Alphas Custom I/O to be implemented in Xilinx XCV405E FPGA (More details in PCI interface/Firmware talks)

10. Bob Hirosky L2  eta Review 26-APR-01 DMA performance is mainly limited by PCI interface The VME-based PIII cards all use PCI 33MHz x 32-bits Max Bandwidth = 132MB/s This rate is explicitly supported by memory controller chips, thus efficient management of MBus Data to PCI feed is the key. We propose using a hardware PCI interface (PLX 9054) considering quanta of DØ data sources and latency times for start of DMA > 100MB/s achievable >~80MB/s w/ periodic DMA interruptions by PIO

11. Bob Hirosky L2  eta Review 26-APR-01 Electrical L2 Alphas power requirements: ~172 W Estimates for L2  eta: ~58W Saving = 104W Or (at 0.10/kWh) = $3000/year for 30 boards

12. Bob Hirosky L2  eta Review 26-APR-01 Collaboration Organization + Device Drivers/Development Env.: UVa 9U + Mezzanine Card Engineering: Orsay Firmware: Maryland + Orsay

13. Bob Hirosky L2  eta Review 26-APR-01 Present Status 1 st 7740 to arrive w/ the week for start of software development Block diagrams for Electrical + Firmware Mechanical prototype work underway Ready to start detailed schematics – deliver prototypes by end of summer