 |
Claims  |
|
|
Having thus described our invention, what we claim as new, and desire to
secure by Letters Patent is:
1. A method of computing the number of units of storage space accessed in a
storage medium, said method comprising the steps of:
setting a predetermined bit in each unit of storage space to a
predetermined state indicative that said unit of storage space has not
been accessed;
setting a counter to a reference count;
accessing a unit of storage space, and if said predetermined bit is in said
predetermined state, changing the state to indicate the unit of storage
space has been accessed;
incrementing said counter in response to the change of state of said
predetermined bit as a result of the unit of storage space being accessed,
with the count of said counter being indicative of the number of units of
storage space accessed; and
determining which one of a plurality of storage management methods to
utilize based on predetermined types of accesses to the storage medium and
the count of said counter.
2. The combination claimed in claim 1, wherein said predetermined types of
accesses to said storage medium comprise a particular file or data base.
3. The combination claim in claim 1, wherein said predetermined types of
accesses to said storage medium comprise accesses by a particular user.
4. The combination claimed in claim 1, wherein said predetermined type of
accesses to said storage medium comprise addresses to a subset of the
address stream for said storage medium.
5. The combination claimed in claim 1, wherein said predetermined type of
accesses to said storage medium comprise access based on a distinct type
of workload.
6. The combination claimed in claim 5, wherein said distinct type of
workload is a query workload.
7. The combination claimed in claim 5, wherein said distinct type of
workload is a sequential workload.
8. A method of computing the number of units of disk storage space
accessed, said method comprising the steps of:
setting a predetermined bit in each unit of disk storage space to a
predetermined state indicative that said unit of disk storage space has
not been accessed;
setting a counter to a reference count;
examining the state of the predetermined bit in a unit of disk storage
space when accessing that unit, and if in said predetermined state,
changing the state to indicate the unit has been accessed; and
incrementing said counter in response to the change of state of said
predetermined bit as a result of the unit being accessed, with the count
of said counter being indicative of the number of units of storage space
accessed;
the method further including the step of:
examining the count of said counter to determine which one of a plurality
of storage management methods to utilize based on predetermined types of
accesses to the storage medium.
9. The method of claim 8, wherein said unit of disk storage comprises a
sector.
10. The method of claim 8, wherein said unit of disk storage comprises a
track.
11. The method of claim 8, wherein said unit of disk storage comprises a
cluster.
12. The method of claim 8, wherein said unit of disk storage comprises a
cylinder.
13. A method of computing the number of units of disk storage space
accessed in a predetermined time interval, said method comprising the
steps of:
counting pulses indicative of units of time by a clock counter for
providing a time count indicative of said predetermined time interval;
setting at least one bit in each unit of disk storage space to a
predetermined state indicative that said unit of disk storage has not been
accessed;
setting an access counter to a preference count;
accessing a unit of disk storage space, and if said at least one bit is in
said predetermined state, changing the state to indicate the unit has been
accessed;
incrementing said access counter in response to the change of state of said
at least one bit as a result of said unit being accessed, with the count
of said access counter being indicative of the number of units of disk
storage space accessed; and
determining the count of said access counter when said clock counter
reaches said time count indicative of said predetermined time interval to
determine the number of units of disk storage space accessed during said
predetermined time interval.
14. The method of claim 13, including the step of:
determining which one of at least two storage management methods to utilize
at a given time, based on predetermined types of accesses to said disk
storage space during said predetermined time interval.
15. The method of claim 14, wherein said unit of disk storage comprises a
sector.
16. The method of claim 14, wherein said unit of disk storage comprises a
track.
17. The method of claim 14, wherein said unit of disk storage comprises a
cluster.
18. The method of claim 14, wherein said unit of disk storage comprises a
cylinder.
19. A method of computing the number of units of disk storage space
accesses of a predetermined type, said method comprising the steps of:
examining storage access requests in an address stream to determine access
requests of said predetermined type;
setting a predetermined bit in each unit of disk storage space to a
predetermined data indicative that said unit of disk storage space has not
been accessed;
setting a counter to a reference count;
accessing a unit of disk storage space in response to an access request of
said given type, and if said predetermined bit is in said predetermined
state, changing the state to indicate the unit has been accessed;
incrementing said counter in response to the change of state of said
predetermined bit as a result of the unit being accessed by an access
request of said predetermined type, with the count of said counter being
indicative of the number of units of storage space accessed; and
invoking a given storage management method in response to said counter
being incremented to a selected count.
20. A method of computing the number of units of disk storage space
accesses of a predetermined type in a predetermined time interval, said
method comprising the steps of:
examining storage access requests in an address stream to determine access
requests of said predetermined type;
counting pulses indicative of units of time by a clock counter for
providing a time count indicative of said predetermined time interval;
setting at least one bit in each unit of disk storage space to a
predetermined state indicative that said unit of disk storage has not been
accessed;
setting an access counter to a reference count;
accessing a unit of disk storage space in response to an access request of
said predetermined type, and if said at least one bit is in said
predetermined state, changing the state to indicate the unit has been
accessed;
incrementing said access counter in response to the change of state of said
at least one bit as a result of said unit being accessed with the count of
said access counter being indicative of the number of units of disk
storage space accessed;
determining the count of said access counter when said clock counter
reaches said time count indicative of said predetermined time interval to
determine the number of units of disk storage space accesses of said
predetermined type during said predetermined time interval; and
invoking a given storage management method in response to said access
counter reaching a selected count during said predetermined time interval.
21. A method of computing the number of units of disk storage space
accesses of at least two different predetermined types in a predetermined
time interval, said method comprising the steps of:
examining storage access requests in an address stream to determine access
requests of said at least two different predetermined types;
counting pulses indicative of units of time by a clock counter for
providing a time count indicative of said predetermined time interval;
setting at least one bit in each unit of disk storage space to a
predetermined state indicative that said unit of disk storage has not been
accessed;
setting an access counter to a reference count;
accessing a unit of storage space in response to an access request of one
of said at least two different types, and if said at least one bit is in
said predetermined state, changing the state to indicate the unit of
storage space has been accessed;
incrementing said access counter in response to the change of state of said
at least one bit as a result of the unit of storage space being accessed,
with the count of said access counter being indicative of the number of
units of disk storage accessed;
determining the count of said access counter when said clock counter
reaches said time count indicative of said predetermined time interval to
determine the number of units of disk storage space accesses of said at
least two different predetermined types during said predetermined time
interval;
invoking a first storage management method in response to said access
counter reaching a first count during said predetermined time interval;
and
invoking a second storage management method in response to said access
counter reaching a second count during said predetermined time interval.
22. The method of claim 21, wherein the first type of access request of
said at least two different predetermined types is based on the user ID
associated with the access request, and the second type of access request
of said at least two different predetermined types is based on the
particular file or database associated with the access request.
23. The method of claim 21, including the step of:
changing at least one of the at least two different types of predetermined
access request parameters.
24. Apparatus for computing the number of units of disk storage space
accesses of a predetermined type, comprising:
a disk storage device;
means for examining storage access requests in an address stream to
determine access requests of said predetermined type;
means for setting a predetermined bit in each unit of disk storage space of
said disk storage device to a predetermined state indicative that said
unit of disk storage space has not been accessed;
a counter which is set to a reference count;
means for accessing a unit of disk storage space in response to an access
request of said given type, and if said predetermined bit is in said
predetermined state, including means for changing the state to indicate
the unit has been accessed;
means for incrementing said counter in response to the change of state of
said predetermined bit as a result of the unit being accessed by an access
request of said predetermined type, with the count of said counter being
indicative of the number of units of storage space accessed; and
means for invoking a given storage management method in response to said
counter being incremented to a selected count.
25. Apparatus for computing the number of units of disk storage space
accesses of a predetermined type in a predetermined time interval,
comprising:
a disk storage device;
means for examining storage access requests in an address stream to
determine access requests of said predetermined type;
counting pulses indicative of units of time by a clock counter for
providing a time count indicative of said predetermined time interval;
means for setting at least one bit in each unit of disk storage space to a
predetermined state indicative that said unit of disk storage of said disk
storage device has not been accessed;
means for setting an access counter to a reference count;
means for accessing a unit of disk storage space in response to an access
request of said predetermined type, and if said at least one bit is in
said predetermined state, changing the state to indicate the unit has been
accessed;
means for incrementing said access counter in response to the change of
state of said at least one bit as a result of said unit being accessed,
with the count of said access counter being indicative of the number of
units of disk storage space accessed;
means for determining the count of said access counter when said clock
counter reaches said time count indicative o said predetermined time
interval to determine the number of units of disk storage space accesses
of said predetermined type during said predetermined time interval; and
means for invoking a given storage management method in response to said
access counter reaching a selected count during said predetermined time
interval. |
|
 |
|
 |
Claims |
|
|
|
Description |
|
|
1. TECHNICAL FIELD
The invention is in the field of disk controllers, and specifically is
directed to computing the number of distinct disk sectors referenced as a
function of time.
2. BACKGROUND ART
Cached disk controllers must manage their caches properly to achieve
optimal performance. The objective is to keep as much active data in the
caches as possible to avoid the very large penalty incurred by having to
access physical disk. There are a number of patents directed to disk
controllers, each having certain advantages and disadvantages.
U.S. Pat. No. 4,101,969 to Lawson et al., is directed to a secondary
storage facility with means for monitoring sector pulses. The problem
solved by this patent is the detection of more than one respondent for a
bus transaction that has a single respondent. When a disk is selected for
reading data, the disk first moves a head to a selected track, then awaits
the selected sector to rotate in position under the head. As the sector
reaches the head, the information that passes under the read head is
filtered and shaped into pulses that are reported over the bus to a host
computer. A problem develops when two or more disk respond to a single
command to read data. This should never occur, but could occur if a device
were to fail, or if noise on the bus were to cause more than one disk to
be selected. Lawson et al describes a mechanism that detects this
condition by observing that two or more drives are attempting to report
pulses over a bus in response to a transaction. There is no teaching
relative to recording the activity of references to a disk or to a
collection of disks.
U.S. Pat. No. 3,577,185 to Belady, is directed to an on-line system for
measuring the efficiency of replacement algorithms in the control and
measurement of a memory hierarchy. Belady measures only faults, that is,
references made to a block of data that is not located in the fast memory
of a hierarchy. Belady determines which faults have occurred that would
have also occurred if an optimal control-algorithm were used instead of
the one actually in use during the measurement period. An optimal
control-algorithm requires perfect knowledge of the future, and thus
cannot be implemented in principle. Belady's invention provides the
information necessary to determine how close a realizable algorithm can
approach the theoretical optimum. Belady records only the number of
faults, and the rate at which they occur. The time at which the faults
occur is not of interest, and thus is not recorded. Information is used to
evaluate a replacement algorithm, and is supposedly for performance
analysis only. There is no suggestions relative to measuring activity of
accesses to a disk.
U.S. Pat. No. 4,542,458 to Kitajima et al., is directed to an attempt at
solving the so-called File Assignment Problem. In other words, it attempts
to place files onto devices (e.g., DASD) in an optimal fashion. It is
assumed that the maximum allowable access rate and maximum allowable
utilizations are known for each device. No methodology is given for
determining these numbers. In the simplest version of the invention, files
are reordered by access rate frequency, and the devices are reordered by
processing speed (i.e., nominal response time estimates). Assume that the
first n-1 files have been assigned by the algorithm to devices (including
the initial case where n=1.) A greedy algorithm then picks, for the nth
file, a `fastest` possible device amongst those devices for which the
access rate and utilization constraints would not be violated. The
constraint equations are then updated, and the process continues with the
n+1st file. Enhancements to this algorithm have additional constraints
such as device failure rates, or response time priorities by type of file
access.
U.S. Pat. No. 4,429,363 to Duke et el. assists in the dynamic operation of
a cache. It suggests the monitoring of cache misses, write operations and
certain status information relating to the last track referenced (LTR). It
then uses this information to help determine when to promote DASD records
to the cache, and occasionally when to demote DASD records from the cache.
As a simple illustrative example, when there are no writes to any track in
a series of requests, the LTR contents are promoted to cache; with a
write, the LTR contents are not promoted. The inhibiting of certain cache
promotions is intended to decrease the cache miss rate for reads. A series
of additional cache promotion/demotion decisions are made for similar
purposes.
According to the present invention, a disk controller computes the number
of distinct sectors referenced as a function of time. The computation is
used by cache management algorithms for optimization of disk cache.
DISCLOSURE OF THE INVENTION
A mechanism within a disk controller computes the number of distinct
sectors referenced as a function of either real or virtual time. This is
termed as "footprint" function which is used by the disk controller as
input for an algorithm that attempts to optimize cache performance.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a block diagram of a footprint module;
FIG. 2 is a block diagram of a system for calculating disk-access
footprints;
FIG. 3 is a general flow chart of the invention;
FIG. 4 is a flow chart indicative of the determination of when to
initialize;
FIG. 5 is a flow chart of the initialization function;
FIG. 6 is a flow chart of the filtering of an address stream;
FIG. 7 is a flow chart of the function of updating the footprint on
filtered sector access;
FIG. 8 is a flow chart of the function of updating a footprint on time
tick; and
FIG. 9 is a flow chart of the function of determining when to change
address filters.
BEST MODE OF CARRYING OUT THE INVENTION
The invention consists of a mechanism that computes the footprint function
by recording access to disk sectors. A disk sector for the purpose of this
invention is a basic unit of information access in a disk system. Each
Write or Read of the disk system specifies a sector in the disk storage
space. For Write operations the information to fill a sector flows to the
disk storage area, and for Read operations, the information stored in a
sector flows from the disk region. In some disk storage systems, larger
regions called cylinders and tracks, each composed of many sectors, can be
Read or Written by individual operations. For purposes of description of
this invention, no distinction is made between sectors, clusters,
cylinders, and tracks, or any other unit of disk storage space. The
footprint function can be defined in terms of any single measure of
storage, and the calculation of the footprint can be generalized directly
to systems that support intermixed access to different units of storage
area simply by treating an access to a composite area as a sequential set
of accesses to the smaller area contained within the composite area.
The footprint function produced by the invention is a sequence of pairs of
numbers of the form (time.sub.i, size.sub.i), which specifies the
footprint function as a function of time. The size of a footprint is the
number of distinct sectors visited since time.sub.o.
To be of maximum benefit for storage management, it may be necessary to
determine the footprint function for accesses to individual data sets or
for access produced by particular processes, quite apart from the
footprint function that describes the behavior of the composite of all
accesses to the entire storage system. Therefore the invention provides a
means for filtering access requests, so as to select specific ones from
the reference stream for processing. Unselected references do not
participate in footprint computation.
Although the footprint function is defined as a function of time, it can
equally be treated as a function of the total number of accesses to disk.
If disk accesses occur at a uniform rate, then the two forms of measuring
footprints produce results that differ only by a scale factor. When time
is used as a basis, the invention requires a real-time clock for supplying
the X-axis information. When access count is used as a basis, the
invention requires an accumulator to count the accumulated accesses. In
what follows, it is assumed that the footprint function is produced as a
function of time, but note that time, total count, or both are equally
easy to produce.
An overview of the invention appears in FIG. 1. This figure shows that the
device that calculates the footprint receives a stream of addresses as one
of its inputs and produces the footprint of the address stream at its
output. The figure shows the output used by a device that manages a disk
cache. This is one possible use for the footprint data. The figure also
shows an input of control information. The control information is
necessary to reset the module when it is first started, and optionally may
include other control functions that govern the behavior of the module as
it gathers footprint data.
FIG. 2 shows the internal structure of the footprint module. This figure
describes the module as it would be constructed using conventional
microprocessor technology, but any implementation that has a control unit,
memory, and arithmetic unit of a conventional form may be utilized, and is
not further elaborated here. Logic blocks 2 and 3, respectively, are the
input and output buffers of a footprint processor 1 through which data are
exchanged with the external modules in the disk system. The address stream
passes through the input buffer 2, and the footprint function calculations
are passed to the output buffer 3 for reporting to other modules.
Logic block 4 is a memory dedicated to the recording of the footprint
function. It can be implemented as a conventional random-access computer
memory. The footprint memory 4 need not be a distinct memory from the
memory contained within the footprint processor 1, provided that
sufficient space is dedicated within the processor memory system to hold
the information required to record the footprint information contained
within an address stream.
Logic block 5 is a filter control unit which uses information to control
the filtering of addresses. The footprint processor 1 is capable of
examining subsets of addresses, and of computing footprints for subsets of
addresses, as well as a footprint for the entire address stream. The
purpose for this computation is to be able to give a separate
characterization of the behavior of each major component of an address
stream. Logic block 5 records the information that is used by the
footprint processor 1 to filter the address stream. It can be a
conventional computer memory, or can be incorporated within the memory
system of the footprint processor 1.
Logic block 6 includes a memory, and holds the clock counter associated
with an address stream. Footprints are calculated as a function of virtual
time. The time base is recorded in logic block 6. Time can be maintained
in any of several ways:
1. Real time, in which case the clock counter is incremented for each tick
of a real-time clock within the logic block,
2. Address-stream reference count, in which case the counter is increased
for each address in the full address stream, and
3. Filtered address-stream reference count, in which case the counter is
increased for each address in the subset of addresses produced by a
filtering operation on the full stream of address.
The clock counter logic block 6 has a counter dedicated to each footprint
function that is being calculated. The module is capable of recording a
multiplicity of footprint functions concurrently, one for each
address-stream filter that is currently active. The logic block can be
implemented as a conventional memory. To increase a counter value, the
clock is retrieved by the footprint processor 1, incremented, and restored
in the counter memory. Alternately, logic block 6 can be implemented as a
collection of counters in which the counter logic is embedded in the logic
block. The footprint processor 1 for such an implementation should have
the capability of resetting a counter, incrementing it, loading an
arbitrary value into the counter, and reading the contents of the counter.
The basic operation of the footprint module is set forth in FIG. 3 which
shows the internal flow of control at the highest level. This figure shows
that as each access reaches the footprint calculation device, the device
in logic block 10 determines if the footprint calculation should be
initialized at this time. In logic block 20, the footprint function is
initialized if the outcome of logic block 10 is a response that forces the
initialization to occur. If logic block 20 is not required, it is skipped.
The next operation in either case is an address-filtering operation
described by logic block 30. Each request is analyzed to determine if the
request participates in the computation of a footprint for some specific
process, or if the request can be ignored. If the request participates, it
is used to update the footprint in logic block 40.
Logic block 50 analyzes the history of accesses to determine if the address
filter should be altered. In this block the invention determines if more
processes need to be analyzed, either because they are new to the system
or because their behavior has not been recently sampled, or for any other
reason relating to the characteristic behavior of the process that
requires a new sampling of the footprint function. One such reason is that
the process behavior may have changed in recent accesses from what has
been the observed past behavior. If the footprint function has to be
gathered for a new process, the address filter is changed.
Another possible change of address filter may be due to the removal of a
process from footprint analysis. This can occur at a fixed time from the
beginning of an analysis, after a fixed number of accesses have been
observed, or after some other similar condition has become true.
FIGS. 4 through 9 show expanded detail of the operation of logic blocks 10,
20, 30, 40 and 50 of FIG. 3.
In FIG. 4 the function of logic block 10 of FIG. 3 is detailed. The
determination of when to initialize a footprint computation is made by
blocks 11, 12, 13 and 14. The purpose of the logic blocks shown in FIG. 4
is to determine if the current characteristics differ sufficiently from
the long-term average characteristics to warrant reinitialization of
footprint gathering. For example, the long-term average may indicate that
a reference stream contains between 10 and 20 distinct process identifiers
in 1-minute intervals of time. The current history indicates that there
are 30 distinct process identifiers in the most recent 1-minute intervals.
This suggests that the characteristics of the disk accesses have changed
recently, and a new footprint should be obtained to describe the current
behavior of the system.
Logic block 11 in FIG. 4 calculates the value of one or more
characteristics of the stream of disk access requests. For the example
given here, block 11 calculates the number of distinct process identifiers
per 1-minute interval, and produces both a short-term running average and
a long-term running average, and the difference in the two. The difference
in the two is an estimate of the stationarity of the disk accesses
characteristics. If current accesses are essentially similar to the
long-term average, then the access requests are stationary over time, and
the value of the stationarity estimator is small. If current accesses
differ substantially from the long-term average, then the access stream
has become nonstationary, and the stationarity estimator indicates this
with a large value. Logic block 12 compares the value of the stationarity
estimator with a threshold. If the value of the estimator is larger than
the threshold, then logic block 13 produces an indication that an
initialization is required, and exits to logic block 20 of FIG. 3. If the
value of the estimator is smaller than the threshold, then logic block 14
produces a normal exit indication, and exits to logic block 30 of FIG. 3.
FIG. 5 shows the initialization process, expanding logic block 20 if FIG.
3. Logic block 21 receives the report that initialization is required from
logic block 13 of FIG. 4. In response to the report, logic block 21 sets
all sector bits to 0. This indicates that the corresponding sectors have
not been accessed. Logic block 22 sets and index i to 0. This index is
used to count the total number of accesses within a footprint. As the
index changes for i to i+1, the footprint time changes time.sub.i to
time.sub.i+1 and associated with the value time.sub.i, the footprint
mechanism captures the value of size.sub.i, the number of distinct sectors
in the footprint accessed up to time time.sub.i.
Logic block 23 resets a clock so that footprints can be measured with a
real-time or virtual-time base, as well as by total number of accesses.
Logic block 24 initializes the footprint count to 0. The footprint count
is the number of unique sectors accessed since initialization.
Footprint functions need not be measured as a function time. It is also
meaningful to measure them as a function of the number of accesses. If
this is the case, the access count must be maintained by the footprint
mechanism. The access count is initialized to zero in logic block 25.
The address filtering mechanism in logic block 30 of FIG. 3 is shown in
FIG. 6. Each of the blocks in this figure are similar. A typical block,
say logic block 31, examines the next access request in an address stream.
If the request satisfies a number of tests that characterizes a particular
footprint measurement, then an exit is taken to the logic block that uses
this request to update the footprint calculation.
As an example of a filter, it may be desirable to calculate the footprint
of all disk accesses produced by a particular user, so the filter examines
the user ID associated with access request. It may also be desirable to
calculate the footprint of accesses to a particular file or database. In
this case, the filter associates each disk access with some larger data
object, and extracts the accesses to a particular data object.
FIG. 6 shows a situation in which each access participates in at most one
filter. If a request is removed by Filter 1 (attribute test #1), for
example, the figure does not show a path by which that request can be
routed to Filter 2 (attribute test #2). It is not essential for the
purposes of the invention that each attribute test be disjoint. It is
permissible for an access to satisfy a plurality of attribute tests and
for an access to be a component of all of the footprint updates whose
filter tests are satisfied by the access. If this is the case, then in
FIG. 6, when an access satisfies a specific filter test, say logic block
31, it used to update the corresponding footprint, and then is returned to
the next logic block, logic block 32 in this case, for additional testing
and processing.
The processing of an access for a specific footprint is illustrated in FIG.
7. If the footprint data is gathered as a function of access count, the
access count is updated in block 41. If the footprint data is gathered as
a function of time, the appropriate means for this update appears below in
the discussion of FIG. 8, and logic block 41 of FIG. 7 is unnecessary.
Continuing the explanation of FIG. 7, it is noted that logic block 42 tests
the sector bit of the accessed sector to determine if this is a new
access. If not, the process is exited. If so, the sector bit is set to 1
to indicate that a subsequent access is not new. This occurs in logic
block 43. The footprint counter is incremented in block 44 to show that
the total size of the footprint has increased.
The processing shown in FIG. 7 is performed for each filter described by
FIG. 6. If a sector access satisfies more than one access filter, the
sector should have distinct sector bits associated with each filter, and
the sector bits should be initialized separately. All sector bits
associated with a specific filter are initialized concurrently, and
different filters can be initialized at different times.
FIG. 8 shows how to capture footprints as a function of time. The figure
shows the processing steps that occur periodically, for example, as forced
by the occurrence of a periodic interrupt. Logic block 45 increments the
clock count, which shows the elapsed time since initialization or since
the last critical time was passed. Logic block 46 tests the counter to
determine if the next critical time has been reached. The purpose of this
block is to sample the footprint function at discrete points in time
rather than to produce the full detail of the footprint function. The
intent is to reduce the amount of data that has to be saved, but the data
reduction also tends to smooth out statistical fluctuations in the data by
averaging the function over a selectable interval of time.
If the current value of the clock count is not a critical time, logic block
46 forces an exit. If the clock count is a critical time, then logic block
47 stores the current value of the footprint counter as the value of
size.sub.i. Logic block 49 calculates the value of the next critical time
at which the footprint function is captured.
If the footprint function is captured as a function of the access number
instead of as a function of time then FIG. 8 can be modified in a
straightforward way for this purpose. Logic block 46 tests an access count
to determine if it is the next critical access count, and Logic block 49
sets the value of the next critical access count. The process depicted in
logic blocks 46,47,48 and 49 in this case are completed immediately after
the processing of logic block 41 of FIG. 7, where it is triggered by the
presence of an access instead of by the presence of a clock interrupt.
The footprint mechanism described thus far has the ability to measure the
footprints for a plurality of access processes as determined by
corresponding filters on the access stream. Moreover, for each filter, the
mechanism has the a | | |
