Advanced Systems Topics - Performance Analysis exercise
-------------------------------------------------------

The accompanying file, Xen_Interrupt_Latencies.csv, contains data
describing a series of (fictional) experiments that compare the
"interrupt latency" in the three versions of the Xen VMM, for varying
numbers of domains (virtual machines) present in the system.

The following questions are based on these data. The data are
presented in comma-separated values format, which is recognised by all
major spreadsheets (OpenOffice, Microsoft Excel, &c.), and which may
easily be parsed in most programming languages. In order to answer the
questions, I recommend that you use a spreadsheet with "pivot table"
or "DataPilot table" functionality.

In submitting your answers, please show all working used to obtain the
answers. This includes, if appropriate, the spreadsheet formulae
and/or program source code used to generate the answers. For
convenience, I would ask that you e-mail your solutions to me at
Derek.Murray@cl.cam.ac.uk.

Derek Murray, 11th March 2007.

------------------------------------------------------------------------

Question 1.

Consider a system with three domains. Based on the provided data, is
there a statistically-significant difference between the interrupt
latency observed under:

(a) Xen 1 and Xen 3?
(b) Xen 2 and Xen 3?
(c) Xen 1 and Xen 2?

------------------------------------------------------------------------

Question 2.

Does varying the number of domains have a statistically-significant
effect on the interrupt latency in:

(a) Xen 1?
(b) Xen 3?

------------------------------------------------------------------------

Question 3.

You must use the entire data set to answer this question.

(a) Do the following changes have a significant effect on the
    interrupt latency:

    (i)   changing the version of Xen?
    (ii)  varying the number of domains?
    (iii) changing the version of Xen and varying the number of
          domains at the same time?

(b) What percentage of the variation in interrupt latency is due to
    the following:

    (i)   changing the version of Xen?
    (ii)  varying the number of domains?
    (iii) the interaction between the version of Xen and the number of
          domains?
    (iv)  measurement error?

(c) Overall, is there a significant difference in interrupt latency
    between:

    (i)   Xen 1 and Xen 3?
    (ii)  Xen 1 and Xen 2?
    (iii) Xen 1 or Xen 2, and Xen 3?

------------------------------------------------------------------------