> For the complete documentation index, see [llms.txt](https://sliu583.gitbook.io/blog/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://sliu583.gitbook.io/blog/conference/index/nsdi-21/distributed-system/ship-computer-or-data-why-not-both.md).

# Ship Computer or Data? Why not both?

![](/files/-MYBmSf2qcobVUbXRpDu)

![](/files/-MYBmbMh6RWDkL1ydPpA)

![](/files/-MYBmhr4lCy4bVtead5J)

![](/files/-MYBn0SO1raGNL-R3hMx)

* Different computation intensity, measure the tput achieved at specific latency target
* How do we know when we don't know underlying workload? 

 

![](/files/-MYBnYpw8ayuasohTuL5)

![](/files/-MYBnpGV2S77b1RhMzdL)

![](/files/-MYBnxFCimxyNUj9Pycc)

![](/files/-MYBo-dYCVKz5Bcie2ZK)

* Rate limiting: latency 
* Request arbitration: optimal fraction of request using ...

![](/files/-MYBoH2uNKZNPP1SC1QF)

![](/files/-MYBoVXmgnnqodfQJ717)

![](/files/-MYBockal1lF0oEkMp8v)

![](/files/-MYBogh4euNiWXZ5TvFg)

![](/files/-MYBotlLBplkMMe_n_kD)

![](/files/-MYBp2nMOa_Emch9z-mq)

* Accommodate different target, but not a combination of targets 
* Ship data 
  * in distributed system, more flexible to deploy 
* Ship code 
  * Cannot re-deploy (interrupt the service) 
  * Run-time changes of the server-extension 
* Partial computation on server side 
  * More fine-grained placement 
  * Limitation: request rate is so high 
    * Overhead? 
* dimension for evaluating data moving cost 

Now:

* cost is the latency, not the network bandwidth