2. PERFORMANCE REPORTING ANALYSIS › 2.4 I/O Configuration Analysis › 2.4.6 DASD Multiple Volume Analysis Inquiry › 2.4.6.2 Technique Tutorial
2.4.6.2 Technique Tutorial
One technique to reduce the cost and floor space requirements
of DASD devices is the implementation of double density
devices. One implementation of this technology is in devices
such as the STC 8650 that allow two 3350 volumes to be mapped
to a single device. Similarly, two 3350 volumes can be
mapped to a single 3380 device, two single density 3380
volumes can be mapped to a single 3380E (double density
3380), or up to three 3380D devices can be mapped to a single
3380K, or up three 3380K devices can be mapped to a single
3390, and so on.
As the name implies, these devices can contain two, or three
times, as much data as a standard device of the same type.
Migration to 3380 devices capable of holding more than one
volume's data involves the physical movement of data residing
on the devices to be mapped together to the 3380 device. The
implementation of double density 3350 type devices such as
the STC 8650 was more complex. To allow the operating system
to use the extra storage space, two apparently unique volume
serial numbers are mapped to the double density device by
specialized utilities provided by the vendor. If STC 8650s
are in use, the operating system believes that it is using
two unique physical devices, while in reality it is using a
single physical device with only one disk arm. In either
implementation, if the volumes that are mapped together have
high I/O activity during the same time period(s), severe arm
contention (i.e., delays) result.
Therefore, you must carefully select the volumes to be mapped
together to ensure that there is not a high probability of
the devices being active during the same time periods. The
algorithm presented in this section allows you to select
volume pairs that have the lowest probability of contention.
If groupings of more than two (3-way or 4-way) volumes are
requested, the algorithm is expanded to reflect this.
To introduce this algorithm, we present the following
problem. Consider the average percent busy values (data
element DVAPCACT) for the six 3380 volumes listed below.
Note that if multiple SYSIDs shared the volumes, the value of
DVAPCACT would be the sum of the values for each SYSID.
These observations represent the activity of the six volumes
for a single hour as reported in the DAYS timespan of the
HARDVA File.
VOLSER DVAPCACT
------ --------
DATA01 3.0
DATA02 4.0
DATA03 1.0
DATA04 7.0
DATA05 2.0
DATA06 5.0
The first step in the process is the calculation of the
contention and summation indices. The contention index is
defined as the product of the device busy percents for all of
the unique volume serial number pairs. Stated as an
equation, the contention index, C, is calculated as:
C(i,j) = DVAPCACT(i) * DVAPCACT(j), for j>i (Equation 10)
The summation index is defined as the sum of the device busy
percents for all of the unique volume serial number pairs.
Stated as an equation, the summation index, S, is calculated
as:
S(i,j) = DVAPCACT(i) + DVAPCACT(j), for j>i (Equation 11)
S(i,j) = 0 if DVAPCACT(i) = 0 or DVAPCACT(j) = 0
Note that the summation index is zero unless both volumes are
active during the hour. Also, most of the elements of the
matrix need not be computed, because the values for (i,j) are
always identical to the values for (j,i).
Figure 2-119 shows the values of the contention and summation
indices for the six volume serial numbers.
| D D D D D D
| A A A A A A
| T T T T T T
C (i,j)| A A A A A A
| 0 0 0 0 0 0
| 1 2 3 4 5 6
-------|--------------------------
DATA01 | . 12 3 21 6 15
|
DATA02 | . . 4 28 8 20
|
DATA03 | . . . 7 2 5
|
DATA04 | . . . . 14 35
|
DATA05 | . . . . . 10
|
DATA06 | . . . . . .
|
S (i,j)|
-------|--------------------------
DATA01 | . 7 4 8 5 8
|
DATA02 | . . 5 11 6 9
|
DATA03 | . . . 8 3 6
|
DATA04 | . . . . 9 12
|
DATA05 | . . . . . 7
|
DATA06 | . . . . . .
Figure 2-119. Contention and Summation Index Values
Based on the contention index values, the best pair of volume
serial numbers to map together would be DATA03 and DATA05,
and the worst pair would be DATA04 and DATA06. However, this
choice is based on data collected for a single hour. During
some other time periods, there might be severe contention
between the volumes.
To evaluate the potential contention between the volumes for
multiple hours, we simply expand the definition of the
contention index to:
C (i,j,k)
| | |
| | |
| | +----- is the hour number, k. It is incremented
| | for every unique CA MICS DAY HOUR pair.
| |
| |
| +------- is the pointer to the second volume serial
| number of the pair.
|
+--------- is the pointer to the first volume serial
number of the pair.
For example, if 20 volumes were studied for 12 hours per day
over a ten-day period, i would range from 1 to 19, j from 2
to 20, and k from 1 to 120.
After we have calculated the values of the contention and
summation indices for all potential pairs for all hours, we
can calculate two additional values for each potential pair.
They are the average contention index and the maximum
summation index. They are defined as:
+- -+
_ K | |
C (i,j) = SUM | C (i,j,k) | / K (Equation 12)
k=1 | |
+- -+
MAXS (i,j) = MAX( S(i,j,k) ) for all k (Equation 13)
where K is the total number of hours studied.
We can then determine volume pairing by selecting the pairs
that have the lowest value of the average contention index.
Since an average can easily hide one or two large values in a
large sample (i.e., a study of many hours), we can use the
maximum summation index as a check value. The inquiry plots
the summation indices for all of the hours studied for any
proposed volume pairs with an average contention index of
greater than 5 or a maximum summation index greater than 25.
For 3350 type volumes, plug-compatible manufacturers have
offered two methods of mapping the volumes onto the double
density 3350 type drive. They are:
o Back-to-Back - Maps the 555 cylinders of each 3350
volume into a contiguous block on the dual density
device. This minimizes seek time as long as most
seeks are within a single volume.
o Interleaved - Interleaves the 555 cylinders of each
3350 volume onto the dual density device; i.e.,
cylinder 1 volume 1, cylinder 1 volume 2, cylinder 2
volume 1, cylinder 2 volume 2, etc. This mapping
technique is recommended for volumes that are expected
to have very light concurrent activity.
We recommend the following selection criteria for choosing
the mapping technique:
o Back-to-Back - For any pair whose average contention
index is less than 50 and maximum summation index is
less than 30.
o Interleaved - For any pair whose average contention
index is greater than 50 and less than 100 and maximum
summation index is less than 50.
Do not map any pair whose contention index is greater than
100 or that has a significant number of hourly summation
indices that are greater than 30 percent (as shown in the
plots).
