Nilesh ranpura systemmodelling

13
Modeling System behaviors ”….A better Paradigm on prototyping -Nilesh Ranpura

Transcript of Nilesh ranpura systemmodelling

Page 1: Nilesh ranpura systemmodelling

“Modeling System behaviors”….A better Paradigm on prototyping

-Nilesh Ranpura

Page 2: Nilesh ranpura systemmodelling

Electronic System Structure

Hardware development

process

Hardware development

process

Software development

process

Software development

process

System development

process

System development

process

Software input/outputHardware input/output

Abstraction trade offs

Page 3: Nilesh ranpura systemmodelling

System Abstraction and Development Process

System Development happens this way Because,1. “Abstraction layers” ensure

threads running concurrently

2. Different approach of specs due to different types of team

3. More than one customer with different nature/features

SPI 5

Design Flow

Concept

Chip/Block specs

Product Specs

System data/environment

Board/proto specs

Architecturespecs

Not documented

Reference Platforms

Hardware and

proto specs

Page 4: Nilesh ranpura systemmodelling

Which are those System Properties ?

Environmental

System PropertiesMixed Signal

Hardware &

Softwareco-exist

Safety/Standards/Compliance

PerformanceNew

Approaches : Green mode Error injection

Page 5: Nilesh ranpura systemmodelling

Processor

FC FC

Fibre Channel Controller

StorageBlade

Compute Blade

PCIeInter-Domain Switch

I/O Hub

CPU CPUMemoryMemory MemoryMemory

PCIe System Interconnect

Switch

LocalStorage

Processor

SATA / SAS Expander

GbE GbE GbE GbE

Dual GbEController

Dual GbEController

I/O Blade

Processor

Compute Blade

I/O Hub

CPU CPUMemoryMemory MemoryMemory

Compute Blade

PCIeInter-Domain Switch

I/O Hub

CPU CPUMemoryMemory MemoryMemory

(1)PCIe Switch Applications – System Interconnect

Page 6: Nilesh ranpura systemmodelling

Model parameter Values

Parameter Value Remark

MPS 128 Testbench

Payload 128/256/512 Testcase

Packet types MRD,MWR,IRD Testcase

Traffic pattern Multicast, One to one, Many to one

Testbench

Speed 2.5Ghz, 5Ghz Testcase

State PM or non PM Testbench

Active Port 2/4/5/6/8 Testcase

No of packets 20/100/500/5000 Testcase

Misc. ECRC, etc

Page 7: Nilesh ranpura systemmodelling

Model Latency definition

Latency is the delay between starting and completing action Latency Definition:

Throughput (pkts/sec) = (total number of pkts(i.e. 500)/(time_t1 - time_t0)) Throughput (bits/sec) = (throughput (pkts/sec) * length * 32)

       In this case, length = Payload size + 3DW headerTheoretical max throughput assumes a 20 byte framing overhead on top of payload.

After removing 8b/10 coding, useful x8 Gen2 unidirectional throughput is 4 GB/s. (4 GBps * payload) / (payload + 20) = theoretical max (second column above)

Payload(Bytes)

Theoretical(GBps) Actual throughput (GBps)Switch efficiency (Actual/Theoretical) %

32 2.462 2.000 81%

64 3.048 2.905 95%

128 3.459 3.360 97%

Page 8: Nilesh ranpura systemmodelling

Usage Model and Error Model

I/O

CPU

NTBP2P

P2P P2P P2P P2P

NTB

I/O I/O I/O I/O

CPU

NTB P2P

P2PP2PP2PP2P

NTB

I/OI/OI/O

.

.

.

Internal Switch Error

I/O1GbE

PCIeSwitch

I/OHub

CPU Memory

.

.

.

I/O1GbE I/O

1GbE

PCIeSwitch

I/OHub

CPU Memory

.

.

.

.

.

.

I/O1GbE

External Error

Page 9: Nilesh ranpura systemmodelling

(2)Modeling Channel properties and Mixed signal for Simple Link

1. Model as much as digital blocks up to last stage2. Last analog Transceiver can be modelled and

converted in to Differential digital by just inverting it.3. Next slide depicts digital noise and

transmission model

Page 10: Nilesh ranpura systemmodelling

(2)Introduce Digital Noise

1. Send 22, 20. 15, 5(which are analog sample’s value) in digital format but in parallal. So no. of data lines = no. of analog samples * 8 bit

2. Introduce noise in numbers by inversion or value changing.

•Inversion•bit stuffing•dummy bits

Noise Model

2220155

43

8 bit value of 20(sample value)

•Inversion•bit stuffing•dummy bits

Noise Model

Page 11: Nilesh ranpura systemmodelling

(2)Actual System

High speed PHY MAC

PAM modulation and Signal path processing block

Page 12: Nilesh ranpura systemmodelling

(2)Actual System

1. Modeled PAM modulation scheme over digital block2. Created noise model to make noise variation between -20dB

to 30dB for high speed signals on Cable.3. Simulated virtual NEXT, FEXT, ISI with predictable noise

model.4. Loop back and system loop back mode tested

High speed PHY MAC

PAM modulation and Signal path processing block

•Inversion•bit stuffing•dummy bits

Noise Model

Page 13: Nilesh ranpura systemmodelling

Thank You, All…!