raww

 INCAST

Spectrum-X switches

Bluefield DPUs

Flow level loadbalancing vs packet level loadbalancing (spraying)

Leads to out-of-order 

NVIDIA's RoCE (RDMA over Converged Ethernet)

RDMA

HPC - high performance computing

Need for - high speed low latency connections 

Eveyday internet use TCP/IP not good for this

Application > OS > TCP/IP stack > NIC card (CPU intence) - adds latency

RDMA Approach

-rNIC offlocading (OS,stack)

1.Infiniband (dedicated nw switch & nic)

2.iWARP - internet wide area RDMS protocol (need iWARP capablie NICs)

3a.RoCE

3b.RoCEv2 (UDP+IP packets)

for RDMA lossless is must

*traditional ethernet is lossy when there is a congestion

1.MTU

2.QoS - prioritse roce packet dscp

3.PFC - priority flow control

   Network switch (receving switch send a pause frame to sending switch , this will make sending switch to stop sending for some time)

   this will help in lossless transaction but intraduce Head-of-line-blocaking 

4.DCQCN - data center quantized congestion notification)

  ECN - Explicit congestion notification

        receving switch sets ECN towards receiver based on the buffer utilization .then receiver generates a special congestion notification packet (CNP) directly back to the sender

and sender slows down the sending rate

Best practice is use both DCQCN + PFC 

AI models

1.opanai(chatcpt)

2.Anthropic(claude)

3.google(gemini)

4.Llama(open-source)

5.Deepseek

we can use browser to access these   > to test prompt

We can use APIs to access these models > to build Apps

Gen AI Framwork (chain,flow,pipeline of components)

1.Langchain

2.LlamaIndex

3.haystack

Building a chain = connecting steps into a pipeline

Type of chains

Simple chain → prompt + LLM

RAG chain → retriever + prompt + LLM

Agent chain → tools + reasoning + LLM

Agentic AI framwork 

1.CrewAI 

2.autogen

3.metaGPT

RAG

RAG sits between your app/agent and the LLM

User

 ↓

Agent / App Logic

 ↓

RAG (data retrieval layer)

 ↓

LLM (reasoning)

 ↓

Response