Service Topology VIP Caching Test

5 servers are used. Two servers, Starlight dtn103 and NEU/MGHPCC sandie-2, act as consumer nodes. Two servers, Starlight dtn098 and Tennessee Tech pegasus act as forwarder nodes with 20 GB cache space each. One server at UCLA acts as the producer node. 

The test has three scenarios: nocache, ARC and VIP. Each consumer node requests 30 4GB files randomly with the zipf(1) distribution for 2 hours.

Multi-path Topology VIP Caching Test

 We use 4 servers in the test. Starlight server (dtn098) acts as a consumer node. Tennessee Tech server acts as a consumer and a forwarder node with 20GB cache capacity.  MGHPCC/NEU server acts as a forwarder node with 20GB cache capacity. UCLA server acts as a producer node. 

The test has three scenarios: round robin with ARC, fast route with ARC, and VIP. Each consumer node requests 30 4GB files randomly with the zipf(1) distribution for 2 hours.


  • Local Test at Caltech SANDIE-7
  • With all data cached in advance in DRAM
  • 18 NDNc consumers with AIMD congestion control
  • 8192 initial window
  • lifetime 500 ms.
  • ~80Gbps


Caltech to Starlight

From Caltech sandie-7(client) to Starlight dtn098(producer), we use 6 forwarding threads in the NDN-DPDK forwarder and cache the requested files at  Starlight dtn098 in advance. On the client side, 18 consumer applications are running with 8192 fixed pipeline sizes each. Each consumer application requests one 1GB file repeatedly for 500 times. Achieve ~40Gbps.

Starlight & SC22 booth to Caltech

  • Test Topology: 2 consumers (SC22 booth, Starlight) and 1 producer(Caltech)
  • 100 Gbps tagged VLAN connections to both consumers.
  • Each NDN-DPDK forwarder uses 6 forwarding threads
  • Each consumer server runs 18 consumer applications to request 18 1GB files named by CMS .
  • Requested data is cached in advance in the DRAM of the Caltech server
  • ~50Gbps

Xrootd Plugin

  • File transfer between Docker containers at the Caltech site
    – A consumer running the plugin and an NDN-DPDK fileserver
    – Same parameters as earlier throughput test (fixed pipeline with size 8192, packet lifetime of 500ms)
    – Result is roughly equivalent to the performance of a single NDNc file transfer client (4.8 Gb/s)

FPGA Accleration

  1. We run NDNcat and then have the forwarder send inputs to both CPU and FPGA for hashing.
  2. Demonstration of the hashing part for a single thread where computation time is compared.

Genomics Data Lake, Kubernetes and NDN


In this demo we will show creating a genomics data lake using K8s over NDN. We publish the data using a file server and can retrieve it using NDNc.

  1. Run file server on K8 cluster on GCP and possibly on NDISE testbed
  2. Run NDNc to retrieve datasets
  3. Demonstrate data can be pulled out of the K8 clusters

Kubernetes compute placement based on names

  1. Create named service endpoints on K8s
  2. Create multiple K8s clusters running the same services 
  3. Show compute requests can be sent to a K8s cluster based on the name
  4. Create multiple service end-points based on different names and show that the network can route requests based on the request names