1
.. This document describes _how_ to do use case analyses and what we want
2
.. to get out of them; for the specific cases see the files referenced by
3
.. performance-roadmap.txt
5
Analysing a specific use case
6
=============================
8
The analysis of a use case needs to provide as outputs:
9
* The functional requirements that the use case has to satisfy.
10
* The file level operations and access patterns that will give the best
12
* A low friction API which will allow the use case to be implemented.
13
* The release of bzr (and thus the supported features) for which the analysis
14
was performed. The feature set of bzr defines the access patterns and data
15
required to implement any use case. So when we add features, their design
16
changes the requirements for the parts of the system they alter, so we need
17
to re-analyse use cases when bzr's feature set changes. If future plans are
18
considered in the analysis with the intention of avoiding rework, these
19
should also be mentioned.
21
Performing the analysis
22
=======================
24
The analysis needs to be able to define the characteristics of the
25
involved disk storage and APIs. That means we need to examine the data
26
required for the operation, in what order it is required, on both the
27
read and write sides, and how that needs to be presented to be
28
consistent with our layering.
30
As a quick example: 'annotation of a file requires the file id looked up
31
from the tree, the basis revision id from the tree, and then the text of
32
that fileid-revisionid pair along with the creating revision id
33
allocated to each line, and the dotted revision number of each of those
34
revision ids.' All three of our key domain objects are involved here,
35
but we haven't defined any characteristics of the api or disk facilities
36
yet. We could then do that by saying something like 'the file-id lookup
37
should degrade gracefully as trees become huge. The tree basis id should
38
be constant time. Retrieval of the annotated text should be roughly
39
constant for any text of the same size regardless of the number of
40
revisions contributing to its content. Mapping of the revision ids to
41
dotted revnos could be done as the text is retrieved, but its completely
42
fine to post-process the annotated text to obtain dotted-revnos.'
44
What factors should be considered?
45
==================================
47
Obviously, those that will make for an extremely fast system :). There
48
are many possible factors, but the ones I think are most interesting to
53
- The time to get bzr ready to begin the use case.
55
- scaling: how does performance change when any of the follow aspects
56
of the system are ratcheted massively up or down:
58
- number of files/dirs/symlinks/subtrees in a tree (both working and
60
- size of any particular file
61
- number of elements within a single directory
63
- number of changes to any file over time
64
(subordinately also the number of merges of the file)
65
- number of commits in the ancestry of a branch
66
(subordinately also the number of merges)
67
- number of revisions in a repository
68
- number of fileids in a repository
69
- number of ghosts in a given graph (revision or per-file)
70
- number of branches in a repository
71
- number of concurrent readers for a tree/branch/repository
72
- number of concurrent writers for objects that support that.
73
- latency to perform file operations (e.g. slow disks, network file systems,
74
our VFS layer and FTP/SFTP/etc)
75
- bandwidth to the disk storage
76
- latency to perform semantic operations (hpss specific)
77
- bandwidth when performing semantic operations.
79
- locality of reference: If an operation requires data that is located
80
within a small region at any point, we often get better performance
81
than with an implementation of the same operation that requires the
82
same amount of data but with a lower locality of reference. Its
83
fairly tricky to add locality of reference after the fact, so I think
84
its worth considering up front.
86
Using these factors, to the annotate example we can add that its
87
reasonable to do two 'semantic' round trips to the local tree, one to
88
the branch object, and two to the repository. In file-operation level
89
measurements, in an ideal world there would be no more than one round
90
trip for each semantic operation. What there must not be is one round
91
trip per revision involved in the revisionid->dotted number mapping, nor
92
per each revision id attributed to a line in the text.
94
Not all the items mentioned above are created equal. The analysis should
95
include the parameters considered and the common case values for each - the
96
optimisation should be around the common cases not around the exceptions.
98
For instance, we have a smart server now; file level operations are relatively
99
low latency and we should use that as the common case. At this point we intend
100
to preserve the performance of the dumb protocol networking, but focus on
101
improving network performance via the smart server and thus escape the
102
file-level operation latency considerations.
104
Many performance problems only become visible when changing the scaling knobs
105
upwards to large trees. On small trees its our baseline performance that drives
106
incremental improvements; on large trees its the amount of processing per item
107
that drives performance. A significant goal therefore is to keep the amount of
108
data to be processed under control. Ideally we can scale in a sublinear fashion
109
for all operations, but we MUST NOT scale even linearly for operations that
110
invoke a latency multiplier. For example, reading a file on disk requires
111
finding the inode for the file, then the block with the data and returning the
112
contents. Due to directory grouping logic we pay a massive price to read files
113
if we do not group the reads of files within the same directory.