From Cmcl

Looking for project partners? Want to discuss ideas? Here's your place.

[edit] Looking for Partners

Name, topics of interest, etc.

• Xin Zhang ( Write to xzhang1@cs. )

 -- Interested in distributed system security, more specifically, 
 -- Wants to explore the the role of trusted computing in securing distributed infrastructures, protocols and services 
    (such as code attestation, network routing and forwarding etc);
 -- Intends to employ and extend the following security framework: 
    TPM (trusted Platform Model) boostraps trusts among remote machines, 
    and secure kernel ensures the integrity of remote process; 
 -- And seeks to answer the following high-level question: by using this trusted computing approach, 
    what are the fundermental benefits and properties that can be achieved, and how can all these affect system design; 
 -- Seems like a design/analysis-oriented project without massive coding and implementation...
 Already has one partner; Also open to other system security/privacy-related topics

• Mikhail Chainani

Interested in storage, distributed systems, file systems, Core OS stuff (VM, scheduling, etc). I'm looking for partners with similar interests. If interested contact me at mchainan@andrew.cmu.edu

• Dongsu Han

Interested in networking, wireless networking. Looking for people with similar interest. dongsuh@cs - got parters, no longer looking for one.

• Adam Wolbach

Looking for 1 or 2 partners on a project related to the Diamond system involving enabling the system to support mobile usage scenarios. The work involves the design and implementation of a new network protocol, among other things. Contact me at awolbach@cs if you're interested.

[edit] Idea discussion

[edit] What are people using their memory for?

Physical RAM is considered a basic resource for modern computer systems, but it is unclear exactly how it is used in many workloads. This project idea stemmed from a discussion of memory and IO sharing in Virtual Machine systems. Specifically, we are interested in the occurrence of sharable pages, i.e. pages whose contents can be found in other pages, either on the same machine, or different machines. Some questions that I would like to answer over the course of this project are:

• How much physical ram in a typical system is sharable?
• How "hot" is sharable memory? If we were to make a graph with time since last access on the x-axis, and number of sharable pages on the y-axis, where would most of the sharable pages fall?
• How do the answers to these questions change if we ignore file-backed pages? anonymous pages? mapped pages? Are most sharable pages hot or cold?
• Ignoring sharing, do anonymous pages tend to be hot or cold? unmapped pages? Most systems make the assumption that anonymous pages are hot, and unmapped file pages are cold. Is this true?

A major component of this project will be finding realistic and interesting workloads to study. It is easy to construct artificial workloads to get any desired answer to these questions. The real concern is what is happening "in the wild."

Another possible direction is to implement techniques for monitoring these attributes in a running system. I have some rough (half-baked?) ideas about implementing some of this monitoring in a very efficient way. These things would be worth exploring, and possibly necessary for the rest of the project. If you are interested in relatively simple kernel hacking, this would be a good project for that. Another, much more difficult, possibility is implementing this monitoring in a VM hypervisor. Some very recent work has explored this (google for "geiger memrx" or "hypervisor exclusive cache"), but it's hardly a solved problem.

If you're interested, contact Jim Cipar, jcipar at andrew ...