Received: from ARTS01.INFN.IT (cosine-gw.infn.it) by optima.CS.Arizona.EDU (5.65c/15) via SMTP
id AA04078; Tue, 29 Jun 1993 08:20:08 MST
Received: From MR(RFCGATEWAY) by MAILER with Id HERMHS 0023081.000741365864
for MAILER@ARTS01.INFN.IT; Tue, 29 JUN 93 17:19 GMT
Message-Id: <HERMHS 0023081.000741365864>
X-Posting-Date: 29-JUN-1993 14:57:44.00
Received: via INFNGW
Date: Tue, 29 JUN 93 17:19 GMT
From: Elias.Eliopoulos@isosun.ariadne-t.gr
Subject:
To: tdbbenchmark@cs.arizona.edu
X-Original-To: tdbbenchmark@cs.arizona.edu
RFC-822-Headers:
Received: by isosun.ariadne-t.gr (4.1/SMI-4.0-MHS-6.0)
id AA23077; Tue, 29 Jun 93 17:57:42 +0300
Dear Rick and Christian,
With respect to the paper on a benchmark on temporal datand Christian,#M#JWith respect to the paper on a benchmark on temporal datshould write something like the following: 'Given that there are many diverse data modelling approaches in the area, each co-
author has tried to identify reasonable queries whose formulation and also their answering should be enabled by a valid time DBMS. This will allow the reader identify from this set of queries those which, in his opinion, are most important and thus perfor
m a personal evaluation of each model, based on two parameters, firstly whether each query can be formulated in some model and, secondly, how easily it can be formulated.'
Secondly, if the 'benchmark' is to appear as a joint paper, we wonder whether it would be wise for every author to express his personal ideas at the end of the paper. In spite of this, our own comments are the following:
Comments on the paper
1. It is unavoidable that the queries and also the schema of the database have been influenced by individual models. For example, in an ungrouped model, the schema could perhaps be completely different, if we were to record the history of employee names.
2. The key and the functional dependencies between various pieces of data are given in the paper. This directly implies that we all agree in 'what the key of a historical relation is', which is not true. We think that we should rather write that 'we provi
de the functional dependencies which we assume that the data satisfy at each time instant'.
3. Many queries (eg 2.1.1, 2.1.2, 2.1.7, 2.1.5) imply that certain scalar and perhaps aggregate functions are supported but we do not specify which they are. In contrast, we are quite specific on the relational operators which are supported.
Minor improvements
Q2.1.5 should be modified as: 'whose salary remained the same for the longer continuous time' or something like this.
Q2.1.7 does not seem to be a reasonable query, in the general case in which there are many managers.
Q2.2.4 should be rephrased as 'For all departments whose managers and budgets have not changed for the last 18 months ...'.
Q2.3.5 is practically identical to Q2.1.5, if we take into consideration the fact that a salary only increases. The distinction will be clearer if in Q2.3.5 we consider a case in which an employee ceases being paid and, after some time he starts being pai
d again with the same salary.
Q2.3.10: Rather than write 'in a department called Toy', better to write 'in the Toy department'. The same remark applies to some other queries.
Q2.4.3 is similar to Q2.4.7. In Q2.4.3 it is better to explicitly write 'at least 5 consecutive years'.
Q4.2.3: Write: 'exceeded', 'continued'.
Q4.2.4: replace 'at' by 'one'. Do you mean exactly one year?
Q4.4.3: Omit the second 'they'.
Q4.10.5: omit 'then'.
New classes of queries
There are many reasonable queries for which no provision has been made in the taxonomy. We distinguish them into groups A, B and C below.
Group A: To simplify the discussion, in the following we do not distinguish between event, interval and element. In addition, we do not distinguish between value, derived and imposed. Instead, we only consider the case that a piece of data d1 is valid at
time t1 and, similarly, d2 is valid at t2. Then the output-based taxonomy may require the output of the following results:
1. t1 t2
2. t1 d2
3. t1 t2 d2
4. d1 t2
5. d1 d2
6. d1 t2 d2
7. t1 d1 t2 d2
Relevant Examples
1. Give the time Edward was in the Toy department and the time his salary became $30K (output of the form t1, t2).
Answer: "2/1/82 - 1/31/87, 6/1/82"
2. Give the time Edward was in the Toy department and the department he is currently working in (output of the form t1, d2).
Answer: "2/1/82 - 1/31/87, Book"
3. Give the time Edward was in the Toy department and his salary history (output of the form t1, t2, d2).
4. Give the department in which Edward was at time 12/31/84 and the time at which his salary became greater than $20K.
Answer: "Toy, 6/1/82"
5. Give Di's salary at time 12/31/85 and 12/31/86.
Answer: "$40K, $50K"
5. Give Di's salary at time 12/31/85 and the department she was in at time 12/31/86. (Note that in this query the time must also be output so as to associate each salary with the respective date).
Answer: "12/31/85, $40K, 12/31/86, Toy"
6. Give Di's salary at time 12/31/85 and her salary history.
Using the above classification of queries, we can see that there are 7 distinct classes of the output-based taxonomy.
Similarly, the top-level selection-based taxonomy might be classified into 7 classes (if reasonable queries can be identified) and this could result in 49 distinct types of queries. If we further consider in conjunction all the cases you have already iden
tified, then it is likely that the number of classes will be further increased
Group B: Queries which require unnestings and nestings (It applies only to grouped models). Some queries are the following.
1. For each department give the current names of the employees who ever worked in it.
Answer: "Toy, ((Edward), (Di))",
"Book, Edward"
2. For each department give the current names of the employees who worked in it and also the respective time.
7. For each time instant within 5/30/82 - 6/2/82, give the distinct salaries which the employees were earning.
Answer: "5/30/82, (($20K), ($30K))",
"5/31/82, (($20K), ($30K))",
"6/1/82, $30K",
"6/2/82, $30K"
8. For every manager give the departments in which he worked and the relevant time.
Answer: "Di, ((Toy, ((1/1/82 - present))))"
9. For every department give his managers and the relevant time.
Answer: "Toy, ((Di, ((1/1/82 - present))))",
"Book, -, -"
Note that this result is different than that of the previous query. In particular a grouped model must be capable of grouping the result of each query in a distinct way as is shown by the pairs of brackets.
10. For every manager list the current names of his employees and the time at which each of them was managed by this particular manager.
Answer: "Di, ((Edward, 2/1/82 - 1/31/87))"
11. For each employee (current name) give his managers and the time at which he was managed by each of them.
Answer: "Edward, ((Di, 2/1/82 - 1/31/87))",
Note that this result is different than that of the previous query. In particular a grouped model must be capable of grouping the result of each query in a distinct way as is shown by the pairs of brackets.
12. List the salary and department of each employee (current name) at times 12/31/84 and 12/31/85 (Here, we want to retrieve data at two distinct time points. Clearly, there is no need to have a nested relation.)
Answer: "12/31/84, Edward, Toy, $30K",
"12/31/84, Di, Toy, $40K",
"12/31/85, Edward, Toy, $40K",
"12/31/85, Di, Toy, $40K"
Clearly, queries like the above can also be classified in a systematic way.
Group C: It includes various queries which we do classify into some particular class.
1. For each employee (current name) list his salary on date 12/31/84, 12/31/85, and 12/31/86.
In this query we want to retrieve employee salaries at more than one specific time instant. It is obvious therefore that the resulting relation must contain for each employee his name and three salaries, each of them recorded next to one of the dates 12/3