1
Analysing a specific use case
2
=============================
4
The analysis of a use case needs to provide as outputs:
5
* The functional requirements that the use case has to satisfy.
6
* The file level operations and access patterns that will give the best
8
* A low friction API which will allow the use case to be implemented.
9
* The release of bzr (and thus the supported features) for which the analysis
10
was performed. The feature set of bzr defines the access patterns and data
11
required to implement any use case. So when we add features, their design
12
changes the requirements for the parts of the system they alter, so we need
13
to re-analyse use cases when bzr's feature set changes. If future plans are
14
considered in the analysis with the intention of avoiding rework, these
15
should also be mentioned.
17
Performing the analysis
18
=======================
20
The analysis needs to be able to define the characteristics of the
21
involved disk storage and APIs. That means we need to examine the data
22
required for the operation, in what order it is required, on both the
23
read and write sides, and how that needs to be presented to be
24
consistent with our layering.
26
As a quick example: 'annotation of a file requires the file id looked up
27
from the tree, the basis revision id from the tree, and then the text of
28
that fileid-revisionid pair along with the creating revision id
29
allocated to each line, and the dotted revision number of each of those
30
revision ids.' All three of our key domain objects are involved here,
31
but we haven't defined any characteristics of the api or disk facilities
32
yet. We could then do that by saying something like 'the file-id lookup
33
should degrade gracefully as trees become huge. The tree basis id should
34
be constant time. Retrieval of the annotated text should be roughly
35
constant for any text of the same size regardless of the number of
36
revisions contributing to its content. Mapping of the revision ids to
37
dotted revnos could be done as the text is retrieved, but its completely
38
fine to post-process the annotated text to obtain dotted-revnos.'
40
What factors should be considered?
41
==================================
43
Obviously, those that will make for an extremely fast system :). There
44
are many possible factors, but the ones I think are most interesting to
49
- The time to get bzr ready to begin the use case.
51
- scaling: how does performance change when any of the follow aspects
52
of the system are ratched massively up or down:
54
- number of files/dirs/symlinks/subtrees in a tree (both working and
56
- size of any particular file
57
- number of elements within a single directory
59
- number of changes to any file over time
60
(subordinately also the number of merges of the file)
61
- number of commits in the ancestry of a branch
62
(subordinately also the number of merges)
63
- number of revisions in a repository
64
- number of fileids in a repository
65
- number of ghosts in a given graph (revision or per-file)
66
- number of branches in a repository
67
- number of concurrent readers for a tree/branch/repository
68
- number of concurrent writers for objects that support that.
69
- latency to perform file operations (e.g. slow disks, network file systems,
70
our VFS layer and FTP/SFTP/etc)
71
- bandwidth to the disk storage
72
- latency to perform semantic operations (hpss specific)
73
- bandwidth when performing semantic operations.
75
- locality of reference: If an operation requires data that is located
76
within a small region at any point, we often get better performance
77
than with an implementation of the same operation that requires the
78
same amount of data but with a lower locality of reference. Its
79
fairly tricky to add locality of reference after the fact, so I think
80
its worth considering up front.
82
Using these factors, to the annotate example we can add that its
83
reasonable to do two 'semantic' round trips to the local tree, one to
84
the branch object, and two to the repository. In file-operation level
85
measurements, in an ideal world there would be no more than one round
86
trip for each semantic operation. What there must not be is one round
87
trip per revision involved in the revisionid->dotted number mapping, nor
88
per each revision id attributed to a line in the text.
90
Not all the items mentioned above are created equal. The analysis should
91
include the parameters considered and the common case values for each - the
92
optimisation should be around the common cases not around the exceptions.
94
For instance, we have a smart server now; file level operations are relatively
95
low latency and we should use that as the common case. At this point we intend
96
to preserve the performance of the dumb protocol networking, but focus on
97
improving network performance via the smart server and thus escape the
98
file-level operation latency considerations.
100
Many performance problems only become visible when changing the scaling knobs
101
upwards to large trees. On small trees its our baseline performance that drives
102
incremental improvements; on large trees its the amount of processing per item
103
that drives performance. A significant goal therefore is to keep the amouont of
104
data to be processed under control. Ideally we can scale in a sublinear fashion
105
for all operations, but we MUST NOT scale even linearly for operations that
106
invoke a latency multiplier. For example, reading a file on disk requires
107
finding the inode for the file, then the block with the data and returning the
108
contents. Due to directory grouping logic we pay a massive price to read files
109
if we do not group the reads of files within the same directory.