|
1
|
Review of GLUE2.1 part 1/2 (J Jensen, UKRI-STFC-RAL)
|
|
2
|
|
|
3
|
|
|
4
|
General comment: in a few places you make reference to grid
|
|
5
|
environments, e.g. in the introduction and the beginning of section
|
|
6
|
6. However, I am wondering if that is too limiting; it might put
|
|
7
|
someone off who is not building a grid. You do mention cloud with the
|
|
8
|
first occurrence, though. Maybe it would be useful for distributed
|
|
9
|
computing more generally.
|
|
10
|
|
|
11
|
As another general comment, some of the descriptions of the schema
|
|
12
|
also touch on how the schema is intended to be used; like in 6.12.1
|
|
13
|
which says that certain Endpoints SHOULD NOT be selected, or 7.10
|
|
14
|
which talks about AppNames. Service (pre-)selection is touched upon
|
|
15
|
several times. There's nothing wrong with that, but it makes it
|
|
16
|
harder to distinguish the normative (Schema) from the non-normative
|
|
17
|
(usage and implementation cf section 1 3rd para).
|
|
18
|
|
|
19
|
|
|
20
|
|
|
21
|
Section 1. Nitpick: renderings are in separate documents? not a
|
|
22
|
separate document (more than one rendering.) There's some repetition
|
|
23
|
as the third para talks about rendering to concrete data models as
|
|
24
|
well.
|
|
25
|
|
|
26
|
Section 5. " .. even though this MAY be measured"? I don't
|
|
27
|
understand.
|
|
28
|
|
|
29
|
What's an "association end"? It's not clear at this stage.
|
|
30
|
|
|
31
|
Section 6.1. Oops. Did you mean "The specialized classes will
|
|
32
|
inherit both the associations and the attributes of the _Entity_
|
|
33
|
class"?
|
|
34
|
|
|
35
|
Section 6.2. What's a Container class? (not mentioned before)
|
|
36
|
|
|
37
|
In 6.3 is the country name free format? There might be a use case for
|
|
38
|
a controlled vocabulary country format (e.g. service selection), even
|
|
39
|
for services spanning countries using ISO3166 (say) - e.g. a service
|
|
40
|
is in "ES,DE,DK,NL,FR,BE" (with no meaning assigned to the ordering.)
|
|
41
|
Name of country could be a bit of a mouthful, like "United Kingdom of
|
|
42
|
Great Britain and Northern Ireland".
|
|
43
|
|
|
44
|
In 6.5.1-2, why is the AdminDomain.ID (resp UserDomain.ID) both * and
|
|
45
|
0..1? - I understand at a handwavy level how domains could participate
|
|
46
|
in other domains, but not how it is modelled in this way when the
|
|
47
|
attributes have the same name (and different multiplicities).
|
|
48
|
|
|
49
|
6.6. I don't understand the "Complexity" and it's not mentioned
|
|
50
|
previously.
|
|
51
|
|
|
52
|
6.7. "OGSA architecture specification" probably needs a reference or a
|
|
53
|
footnote.
|
|
54
|
|
|
55
|
I seem to remember there is an equivalent of WSDL for REST web
|
|
56
|
services but I can't remember what it's called. WSDL is specific to
|
|
57
|
SOAP?
|
|
58
|
|
|
59
|
6.12 in the third para, what defines a set of policies? It is hard to
|
|
60
|
parse this paragraph.
|
|
61
|
|
|
62
|
7.1 For the last Para I suggest adding a diagram showing the two
|
|
63
|
situations - again the para requires some patience to parse. But
|
|
64
|
unlike the one in 6.12, it does make sense.
|
|
65
|
|
|
66
|
7.2 what is a measurable attribute?
|
|
67
|
|
|
68
|
If by Quality_Level you mean TRL, there are quite a few definitions.
|
|
69
|
|
|
70
|
7.3 I find it slightly unhelpful that some memory units (RAM) are in
|
|
71
|
MB, and others (disk) are in GB. I know that disks usually have more
|
|
72
|
capacity than RAM but given the blurring of the boundary between them
|
|
73
|
with NVM, fast disk, slow cache, etc., it seems unhelpful.
|
|
74
|
|
|
75
|
7.4 what's an "accelerator slot"? Is it how many jobs can share an
|
|
76
|
accelerator? Or how many jobs can share all accelerators? For
|
|
77
|
example, on a large GPU system, we would typically have at most one
|
|
78
|
job per GPU but there will be several GPUs. The matching system might
|
|
79
|
look for a resource where its job can get a dedicated GPU. I guess
|
|
80
|
my point is similar to your discussion of the working area allocation
|
|
81
|
in Table 1.
|
|
82
|
|
|
83
|
On 7.8, I noted that sometimes an execution environment could have a
|
|
84
|
non-integer number of CPUs, like if it runs in a container with
|
|
85
|
restricted CPU. Not everything in life is HPC...
|
|
86
|
|
|
87
|
8.1 "weak aggregation"?
|
|
88
|
|
|
89
|
End of 8.1 eek- this is the same as the 7.1 - diagram would be
|
|
90
|
helpful.
|
|
91
|
|
|
92
|
|
|
93
|
|
|
94
|
|
|
95
|
Language nitpicks:
|
|
96
|
|
|
97
|
enables to link -> enables linking
|
|
98
|
easy of reading -> ease of reading
|
|
99
|
|
|
100
|
"the value of zero means that it is allowed to refrain from publishing
|
|
101
|
a value for the related attribute" - it would be more readable to say
|
|
102
|
"it is allowed to omit the attribute" - and later that sentence, "even
|
|
103
|
though this MAY be measured" - what does this mean?
|
|
104
|
|
|
105
|
For "global unique id" I'd suggest "globally unique id"
|
|
106
|
|
|
107
|
6.5.2 inconsistent spelling "A User Domain" and "An User Domain"
|
|
108
|
|
|
109
|
6.12 "We RECOMMEND to not represent the association [..] explicitly"
|
|
110
|
-> "the association [..] SHOULD NOT be represented explicitly" or
|
|
111
|
"representing the association [..] explicitly is NOT RECOMMENDED"
|
|
112
|
(using SHOULD NOT would be better)
|
|
113
|
|
|
114
|
Section 8. "leverages on" -> "leverages" (if you *must* use that
|
|
115
|
word.)
|
|
116
|
|
|
117
|
Also section 8, "bed in Chapter 7" ????
|
|
118
|
|