DTIC ADA465532: A New Approach to Real-Time Transaction Scheduling

7 Pages·0.65 MB·English

by  Defense Technical Information Center

A New Approach to Real-Time Transaction Scheduling
Sang H. Son and Juhnyoung Lee
Department of Computer Science
University of Virginia
Charlottesville, Virginia 22903, USA
[email protected], FAX: [phone]
Abstract the operations in order to avoid aborting them later.
However, the very idea of delaying an operation is
A real-time database system differs from a conventional opposed to real-time systems. Besides, the degree of
database system because in addition to the consistency concurrency is low in the two-phase locking based algo-
constraints of the database, timing constraints of indivi- rithms because concurrent read and write locks on the
dual transaction need to be satisfied. Various real-time same data object by seveiral transactions are not possible.
transaction scheduling algorithms have been proposed Furthermore, two-phase locking has some inherent prob-
which employ either a pessimistic or an optimistic lems such as the possibility of deadlocks and unpredict-
approach to concurrency control. In this paper, we able blocking times, which are serious problems for
present new real-time transaction scheduling algorithms real-time concurrency control.
which employ a hybrid approach, Le., a combination of
An optimistic approach is a natural alternative. An
both pessimistic and optimistic approaches. These proto-
optimistic scheduler aggressively schedules all opera-
cols make use of a new conflict resolution scheme called
tions, hoping that nothing will go wrong, such as a non-
dynamic adjustment of serialization order, which sup-
serializable execution. .Later when the transaction is
ports priority-driven scheduling, and avoids unnecessary
ready to commit, conflicts are checked for and a conflict
aborts. Our experimental results indicate that hybrid pro-
resolution scheme is then applied to resolve the conflicts,
tocols outperform other real-time concurrency control
if any. Ideally an optimistic approach has the properties
protocols in certain performance metrics.
of non-blocking and deadlock freedom, which make it
suitable to real-time transaction processing. In addition,
Key words: real-time database, scheduling, concurrency
it has a potential for high degree of parallelism. How-
control, deadline
ever, the use of aborting for conflict resolution in
optimistic schedulers results in a problem that transac-
tions can end up aborting after having paid for most of
the transaction’s execution. This problem can particu-
1. Introduction larly be serious for real-time transaction scheduling,
A real-time database system (RTDBS) differs where timing constraints of transactions should be
from a conventional database system because in addition satisfied. For conventional database systems, it has been
to the consistency constraints of the database, timing shown that optimal performance can be achieved by
constraints of individual transaction need to be satisfied. combining blocking and aborting. We expect the same
In order to provide real-time response for queries and with RTDBS.
updates while maintaining the consistency of data, real- In this paper, we present two hybrid real-time con-
time concurrency control should involve efficient
currency control protocells which combine pessimistic
integration of the ideas from both database concurrency and optimistic approaches to concurrency control in
control and real-time scheduling. Various real-time con-
order to control blocking and aborting in a more effective
currency control protocols have been proposed which manner. We compare the performance of these new pro-
employ either a pessimistic or an optimistic approach to
tocols with that of other real-time concurrency control
concurrency control. protocols proposed.
Most of the initial work in real-time concurrency The first protocol is based on a priority-based lock-
control has been conducted on utilizing two-phase lock- ing mechanism to support real-time scheduling by adjust-
ing as the base of real-time concurrency control. The ing the serialization order dynamically in favor of high
two-phase locking is termed as being pessimistic,
priority transactions. This protocol uses the phase-
because it, in anticipation of data conflicts, tends to delay dependent control of optimistic approach to support
This work was supported in part by ONR. by NOSC, and by IBM.
I77
0-8186-28154192$ 3.00 Q 1992IEEE
Report Documentation Page Form Approved
OMB No. 0704-0188
Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and
maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington
VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it

does not display a currently valid OMB control number.
1. REPORT DATE 3. DATES COVERED
1992 2. REPORT TYPE 00-00-1992 to 00-00-1992
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
A New Approach to Real-Time Transaction Scheduling
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION
University of Virginia,Department of Computer Science,151 Engineer’s REPORT NUMBER
Way,Charlottesville,VA,22904-4740
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)
11. SPONSOR/MONITOR’S REPORT
NUMBER(S)
12. DISTRIBUTION/AVAILABILITY STATEMENT
Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES
14. ABSTRACT
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF
ABSTRACT OF PAGES RESPONSIBLE PERSON

a. REPORT b. ABSTRACT c. THIS PAGE 6
unclassified unclassified unclassified
Standard Form 298 (Rev. 8-98)
Prescribed by ANSI Std Z39-18
dynamic adjustment of serialization order. Recently, real-time concurrency control protocols
based on optimistic method have been proposed and stu-
The second protocol is a combination of optimistic
died [Har90, Hua91, Lin90, C00911. Haritsa et d.
concurrency control and timesatmp ordering. This proto-
[Ha1901 proposed a group of optimistic real-time con-
col also uses the phase-dependent control of optimistic
currency control protocols and evaluated them on a simu-
approach. Furthermore, this protocol employs the notion
lation model. They have also conducted a study on the
of dynamic timestamp allocation and dynamic adjust-
relative performance of locking-based protocols and
ment of serialization order using timestamp interval
optimistic protocols, and concluded that optimistic con-
[Bok871, with which the ability of early detection and
currency control protocols outperform two-phase
resolution of nonserializable execution is improved, and
locking-based protocols over a wide range of system util-
unnecessary aborts are avoided.
ization. Huang et al. [Hua91] also conducted a similar
performance study of real-time optimistic concurrency
2. Related Work
control protocols, but on a testbed system, not through
A real-time database system (RTDBS) is a tran- simulation. They examined the overall effects and the
saction processing system designed to handle workloads impact of the overheads involved in implementing real-
where transactions have completion deadlines. The time optimistic concurrency control on the testbed sys-
objective of such system is to meet these deadlines. The tem. Their experimental results contrast with the results
real-time performance of an RTDBS depends on several in EHar901, showing that optimistic concurrency control
factors such as the database system organization, the may not always outperform a two-phase locking-based
underlying processors and disk speeds. For a given sys- protocol which aborts the lower priority transaction when
tem configuration, however, the primary determinants for a conflict occurs. They pointed out the fact that the phy-
the real-time performance are the policies used for sical implementation schemes have a significant impact
scheduling transaction accesses to system resources, on the performance of real-time optimistic concurrency
because these policies determine when service is pro- control.
vided to a transaction.
Recently, research on the scheduling problem in 3. Hybrid Approach
real-time database systems has been active [Abb88, Concurrency control protocols induce a serializa-
Buc89, coo91, W O , H ua90, Hua91, Lin90, Sha91,
tion order among conflicting transactions. For a con-
Son90bl. As mentioned earlier, most of the current real-
currency control protocol to accommodate the timing
time concurrency control schemes are based on two-
constraints of transactions, the serialization order it pro-
phase locking [Abb88, Hua90, Sha911. Abbott and duces should reflect the priority of transactions. How-
Garcia-Molina [Abb88] described a group of lock-based ever, this is often hindered by the past execution history
real-time concurrency control schemes for scheduling
of transactions. A higher priority transaction may have
soft real-time transactions, and evaluated the perfor-
no way to precede a lower priority transaction in the seri-
mance of those protocols through simulation.
alization order due to previous conflicts. For example,
Some inherent problems of two-phase locking such TH and TL are two transactions with TH having a higher
as the possibility of deadlocks and unpredictable block- priority. If TL writes a data object x before TH read it,
ing times are serious for real-time transaction scheduling. then the serialization order between TH and TL is deter-
In addition, a priority inversion can occur when a lower mined as TL + TH. TH can never precede TL in the seri-
priority transaction blocks the execution of a higher alization order as long as both reside in the execution his-
priority transaction. Sha et al. [Sha91] proposed a con- tory. Most of the current (real-time) concurrency control
servative real-time concurrency control scheme called protocols resolve this conflict either by blocking TH until
priority ceiling, which prevents deadlocks and transitive TL releases the writelock or by aborting TL in favor of the
blocking. higher priority transaction TH. Blocking of a higher
priority transaction due to a lower priority transaction is
Huang et al. [Hua90] developed and evaluated a
contrary to the requirement of real-time scheduling.
group of real-time protocols for handling CPU schedul-
Aborting is also not desirable because it degrades the
ing, data conflict resolution, deadlock resolution, aansac-
system performance and may lead to violations of timing
tion wakeup, and transaction restart. Their study is based
constraints. Furthermore, some aborts can be useless
not on simulation but by actual implementation on a
when the transaction which caused the abort is aborted
testbed system called RT-CARAT. They concluded that
due to another conflict. The objective of our hybrid
CPU scheduling protocols have a significant impact on
approach is to avoid such unnecessary blocking and
the performance of RTDBS, that they dominate all other
aborting.
protocols, and that the overhead incurred in locking is
non-negligible and cannot be ignored in real-time con-
currency control analysis.
178
3.1. Integrated Real-Time Locking Protocol (1) rTH [XI, PwTL [XI
The resulting serialization order is TH -+ TL, hence
The first protocol, Integrated Real-Time Locking
(IRTL), combines locking and optimistic concurrency satisfies the priority order, and does not need to adjust the
serialization order.
control. By using a priority-dependent locking protocol,
the serialization order of active transactions is adjusted (2) pwTH [x 1, rTL [x1
dynamically, making it possible for transactions with Either of the two serialization orders can be induced with
higher priority to be executed first so that higher priority this conflict; TL -+ TH with immediate reading, and
transactions are never blocked by uncommitted lower TH 4 TL with delayed reading. Certainly, the latter
priority transactions, while lower priority transactions should be chosen for priority scheduling. The delayed
may not have to be aborted even in the face of a conflict. reading in this protocol rneans blocking of rTL[ XI by the
The adjustment of the serialization order can be con-
writelock of Tu on x.
sidered as a mechanism to support real-time scheduling.
(3) rTL [XI, PWT, [XI
This protocol is an integrated protocol because it
If TL is in read phase, abort TL. If TL is in its wait phase,
uses different solutions for reaawrite (rw) and
avoid aborting TL until TH commits in the hope that TL
write/write (ww) synchronization, and integrates the
gets a chance to commit before TH does. If TH commits,
solutions to the two subproblems to yield a solution to
TL is aborted. But if TH is aborted by some other
the entire problem.
conflicting transaction, then TL is committed. With this
The protocol is similar to optimistic concurrency policy, we can avoid unnecessary and useless aborts,
control in the sense that each transaction has three while satisfying priority scheduling.
phases, but unlike the optimistic method, there is no vali-
dation phase. This protocol’s three phases are read, wait, (4) pwTL [XI, rTH [XI
Either of the two Serialization orders can be induced with
and write phases. The read phase is similar to that of
this conflict; TH -+ TL with immediate reading, and
optimistic concurrency control wherein a transaction
reads from the database and writes to its local TL -+ TH with delayed reading. If TL is in its write
phase, delaying TH is the only choice. This blocking is
workspace. In this phase, however, conflicts are also
not a serious problem far TH because TL is expected to
resolved by using transaction priority. While other
finish writing x soon. TH can read x as soon as TL
optimistic real-time concurrency control protocols
finishes writing x in the database, not necessarily after TL
resolve conflicts in the validation phase, this protocol
completes the whole write phase. If TL is in its read or
resolves them in the read phase. In the wait phase, a
wait phase, choose immediate reading.
transaction waits for its chance to commit. Finally, in the
write phase, updates are made permanent to the database. As transactions are being executed and conflicting
operations occur, all the information pertaining to the
(1) Read phase
induced dependencies in the serialization order needs to
This is the normal execution of a transaction be retained. In order tci maintain this information, we
except that all writes are on private data copies in the associate the following with each transaction; two sets,
local workspace of the transaction instead of on data before-trans-set and after-trans-set, and a count, before-
objects in the database. Such write operations are called count R. The before-trans-set (respectively, after-trans-
prewritcs. The prewrites are useful when a transaction is set) of a transaction contains all the active lower priority
aborted, in which case the data in the local workspace is transactions that must precede (respectively, follows) this
simply discarded. No rollback is required. transaction in the serialization order. The before-count
In this phase read-prewrite and prewrite-read of a transaction is the number of the higher priority tran-
conflicts are resolved using a priority based locking pro- sactions that precede thiis transaction in the serialization
tocol. A transaction must obtain the corresponding lock order. When a conflict occurs between two transactions,
before it reads or prewrites. According to the priority their dependency is determined and then the values of
locking protocol, higher priority transactions must com- their before-trans-set, after-trans-set, and before-count
plete before lower priority transactions. If a low priority are changed accordingly.
transaction does complete before a high-priority transac- (2) Wait Phase
tion, it is required to wait until it is sure that its commit-
The wait phase allows a transaction to wait until it
ment will not lead to the higher priority transaction being
can commit. A transaction in the wait phase can commit
aborted.
if all transactions with higher priority that must precede it
Suppose Tu and TL are two active transactions and in the serialization orider, are either committed or
TH has higher priority than TL, there are four possible aborted. Since the before-count of a transaction is the
conflicts as follows. number of such transactions, the transaction can commit
only if its before-count becomes zero.
I79
A transaction in the wait phase may be aborted due transaction is chosen. These provide a sufficient condi-
eo two reasons; if a higher priority transaction requests a tion for serialization. While satisfying (Cl) and (C2), the
conflicting lock or if a higher priority transaction that protocol also adjusts the serialization order in favor of
must follow this transaction in the serialization order priority order without violating data consistency.
commits.
Most timestamp-based concurrency control proto-
Once a transaction in its wait phase finds a chance cols use static timestamp allocation scheme, i.e., each
to commit, then it commits and switches to its write transaction is assigned a timestamp value at its startup
phase and releases all readlocks. The transaction is time, and a total ordering instead of a partial ordering is
assigned a final timestamp which is the absolute seriali- built up. This total ordering does not reflect any actual
zation order. conflict. Hence, it is possible that a transaction is aborted
even when it requests its first data access. Besides, the
(3) Write Phase
total ordering of all transactions is too restrictive, and
In the write phase, the transaction is considered to degrades the degree of concurrency considerably. With
be committed. All committed transactions are serialized dynamic timestamp allocation, serialization order among
by the final timestamp order. Updates are made per- transactions are dynamically constructed on demand
manent to the database while applying Thomas’ Write whenever actual conflicts are occurring. Only necessary
Rule (TWR) for write-write conflicts [Ber87]. After partial ordering among transactions is constructed instead
each operation the corresponding writelock is released. of a total ordering from the static timestamp allocation.
With the dynamic adjustment of serialization order Furthermore, in this protocol, the use of timestamp
of this protocol, priority order of transactions are retained interval instead of a single value for timestamp allows
as much as possible, which promises more timing con- more flexibility to adjust serialization order. Subse-
straints of real-time transactions to be satisfied. In addi- quently, this protocol can provide a high degree of paral-
tion unnecessary and useless aborts are avoided as much lelism, and avoid many unnecessary aborts. As other
as possible. This protocol also takes advantage of the optimistic protocols, this protocol divides the execution
potential for high degree of parallelism of optimistic con- of a transaction into three phases: read, validation, and
currency control, because concurrent read and write write. However, unlike other optimistic protocols,
locks are possible. This protocol is proved to be conflicts and nonserializable executions are detected and
deadlock-free [Lin901. resolved during the read phase. This early resolution of
nonserializable execution is another advantage of this
3.2. An Optimistic Method with Timestamp Inter- protocol. The main weakness of this protocol is that
vals priority inversion still can occur, because this protocol
The second protocol is a combination of optimistic does not involve transaction priority for scheduling deci-
concurrency control and dynamic timestamp allocation sion.
scheme [Bok87]. This protocol constructs the serializa- The algorithm of the protocol described so far is
tion order dynamically by using timestamp interval, asso- the basic form of the protocol, which does not use tran-
ciated with every active transaction. It updates the times- saction priority in scheduling decision. Not only the pro-
tamp intervals of active transactions to adjust their serial- tocol in its basic form has several advantages over other
ization order. optimistic protocols, but also it can be improved in
This protocol enforces senalizability by satisfying several ways by considering priority scheduling. One
the following two conditions through every read, possibility is to manipulate the size of timestamp inter-
prewrite, and validation; vals of active transactions according to their priority.
Because the size indicates the amount of possibility of
(Cl) Each timestamp interval constructed when a tran- restarting the transaction, a transaction with higher prior-
saction accesses a data object should preserve the ity needs to have a larger timestamp interval than a tran-
order induced by the timestamps of all committed saction with lower priority. This strategy of manipulat-
transactions which have also accessed that data ing the size of timestamp intervals depends on the policy
object. of choosing final timestamp of the validating transacuon
from its timestamp interval. When choosing the final
(C2) The order induced by final timestamp of a validat-
timestamp for a validating transaction, the protocol
ing transaction should not destroy the serialization
should check the priority of its conflicting transactions,
order constructed by the past execution, i.e., by
and decide the timestamp in such a way that higher prior-
committed transactions.
ity transactions are left with larger timestamp intervals.
(Cl) is used in the read phase of a transaction, when it We have developed several extensions of the algorithm
reads or prewrites a data object. (C2) is for validation regarding this final timestamp selection. In addition,
phase, when a final timestamp for the validating because this protocol is based on optimistic control using
forward validation mechanism, it can also be extended to resulting in more restarts. Finally, a validating transac-
support real-time scheduling by combining priority abort tion with little slack time sometimes has to wait for a
and priority sacrifice mechanisms with the validation conflicting high priority transaction which is still in the
mechanism, as in OPT-ABORT, OPT-SACRIFICE and early stage of its read phase, because the protocol blocks
OPT-WAIT protocols proposed in [Hadlo]. the validating transaction ,without considering the current
stages of the conflicting transactions. This again will
4. Performance Evaluation decrease the chance of meeting transaction deadlines.
We have conducted a comparative performance Another observation is that OPT-SACRIFICE per-
evaluation of various real-time concurrency control forms only slightly better than OPT-CMT. The benefit
schemes, using a database prototyping tool [Son90]. We obtained by using priority information in OPT-
have examined one pessimistic protocol called High SACRIFICE has been almost nullified by two potential
Priority (HP) protocol [Abb88], which is based on the problems of OPT-SACRIFICE; wasted sacrifice and
two-phase locking, and employs priority abort policy for mutual sacrifice [Har90].
conflict resolution. We have included three optimistic
Finally, the experimlental results shows the hybrid
protocols; OPT-CMT, OPT-SACRIFICE, and OPT-
protocol, RTL performs competitively. In particular,
WAIT [Har90, Hua911. Under OPT-CMT, the validating
under low data contention, it outperforms all of the other
transaction always commit, while aborting all the
real-time concurrency control. The degraded perfor-
conflicting active transactions. OPT-S ACRIFICE uses mance of IRTL as data cointention increases is primarily
priority abort policy for conflict resolution, while OPT-
due to the overhead incurred in maintaining the informa-
WAIT employs priority wait policy. As a hybrid proto- tion on the induced dependencies in the serialization
col, we have also included IRTL. The primary metric
order among transactions. We have to pay serialization
used for performance evaluation is percentage of missed
order information retaining overhead to take advantage
deadlines, which is the percentage of transactions that
of an intelligent conflict resolution policy called dynamic
are not completed by their deadline. Furthermore, tran- adjustment of serialization order, which makes it possible
sactions are divided into ten priority groups according to to avoid unnecessary bloclting and aborting. In terms of
their priorities. We have also employed deadline missing the discriminating power for different priority groups,
percentage with respect to each priority group, as one of IRTL and OPT-ABORT are good, and OPT-CMT is the
our performance metrics. It identifies more precisely the
worst.
discriminating power of the real-time concurrency con-
Figure 1 shows the percentage of missed transac-
trol schemes in question.
tion deadlines for five prcltocols; HP,O PT-CMT, OPT-
Our experimental results show that over the entire
WAIT, OPT-SACRIFICE, and IRTL under different data
operational range, optimistic schemes outperform the contention levels by varying transaction interarrival time.
locking-based pessimistic protocol, HP. This result
Figure 2 shows the miss percentage for the same five
agrees with the results from [Har90]. The use of block- schemes under different levels of data contention and
ing in pessimistic control may incur more performance
system load by varying the transaction size, while fixing
degradation than the use of aborting in optimistic control the values of other parameters. Figure 3 shows the per-
in real-time database systems. Our experimental results centage of missed deadlines with respect to priority
provide a solid ground for this intuitive reasoning for the
group over four protocols: OPT-CMT, OPT-
use of aborting against blocking in real-time concurrency SACRIFICE, OPT-WAIT, and IRTL. The value of the
conuol. mean interarrival time in Figure 3 is 50 msec, which
Optimistic concurrency control protocols based on represents a normal data contention.
aborting such as OPT-CMT and OPT-SACRIFICE per-
form better than optimistic protocols based on blocking
such as OPT-WAIT. Contrary to the previous result in REFERENCES
[Har90], OPT-WAIT performs even worse than OPT- [Abb88] R. Abbott, H. Garcia-Molina, "Scheduling
CMT which does not use transaction priority informa-
Real-time Transactions: A Performance
tion. We may attribute the degraded performance of
Evaluation," Proceedings of the 14th VLDB
OPT-WAIT to its use of blocking, resulting in several
Conference, 1988.
potential problems. First, if a transaction finally commits
[Ber87] Bernstein, P. A., V. Hadzilacos, and N.G ood-
after waiting for sometime, it causes all its conflicting
man, Concurreircy Control and Recovery in
transactions with lower priority to be restarted at a later
Database System, Addison-Wesley, Reading,
point in time, hence decreasing the chance of these tran-
Mass., 1987.
sactions meeting their deadlines. Second, while a vali-
dating transaction waits, new conflicts can occur and the [Bok871 Boksenbaum, C., M. Cart, J. Feme, and J.
number of conflicting transactions is increased, hence Pons, "Concurrent Certifications by Intervals
of Timestamps in Distributed Database Sys-
tems," IEEE Transactions on Software
Engineering, Vol. SE-13, NO. 4, April 1987. 100
[Buc89] Buchmann, A. et al., "Time-Critical Database OPT-CnT C
Scheduling: A Framework for Integrating OPT-SAOCPTR-IWFAICITE +-+-
Real-Time Scheduling and Concurrency Con- BO IRTHLP -L*-
trol," 5th Data Engineering Conference,
February 1989. 60
Cook, R. P., S. H. Son, H. Y. Oh, and J. Lee,
J
"New Paradigms for Real-Time Database
40
Systems," 8th IEEE workshop on Real-Time
Operating Systems and Software (in Conjunc-
tion with) IFACIIFIP Workshop on Real-Time 20
Programming, May 1991.
mar901 J. R. Haritsa, M. J. Carey, and M. Livny, 0
10 20 30 40 50 60 10 80 30
"Dynamic Real-Time Optimistic Concurrency m9.n I"t.r.rllY.1 tin*
Figure 1. Mean interarrival time
Control," IEEE Real-Time Systems Sympo-
sium, Orlando, Florida, December 1990.
[Hua90] Huang, J., J. A. Stankovic, D. Towsley, and
100 1
K. Ramamritham, "Real-Time Transaction
Processing: Design, Implementation, and Per-
formance Evaluation," Univ. of Mas-
sachusetts, COINS Technical Report 90-43,
May, 1990.
[Hua91] J. Huang, J. A. Stankovic, K. Ramamritham,
and D. Towsley, "Experimental Evaluation of
Real-Time Optimistic Concurrency Control
Schemes," VLDB Conference, Barcelona,
Spain, Sept. 1991.
[Lin90] Y. Lin and S. H. Son, "concurrency Control
in Real-Time Database Systems by Dynamic
Adjustment of Serialization Order," IEEE
Real-Time Systems Symposium, Orlando,
Figure 2. Transaction size
Florida, December 1990.
[Shag11 Sha, L., R. Rajkumar, S. Son and C. Chang,
"A Real-Time Locking Protocol," IEEE
100
Trans. on Computers, Vol. 6, No. 7, July
1991.
.
[Son901 S. H. Son, "An Environment for Prototyping 80
Real-Time Distributed Databases," tnterna-
tional Conference on Systems Integration, t
Momstown, New Jersey, April 1990, pp c7
358-367.
40
[Son90bl Son, S. H., and I. Lee, "Scheduling Real-
Time Transactions in Distributed Database
Systems," 7th IEEE Workshop on Real-Time
Operating Systems and Software, Charlottes-
ville, Virginia, May 1990.
[Yu90] Yu, P., and D. Dias, "Concurrency Control
Using Locking with Deferred Blocking," 6th Figure 3. Miss percentage per priority group
Intl. Conf Data Engineering, Los Angels,
Feb. 1990, pp. 30-36.
182