|
Performance-related term
|
Definition
|
|
Pool
|
The storage pool of a storage system. A pool is a logical grouping of one or more physical disk arrays or external virtualized storage, which provide storage capacity to the pool.
Pool attributes vary by storage-system type, but generally define the capabilities of the host-accessible storage that a user can create from that pool. For performance modeling, a
pool is associated with a workload. A workload can have at most one pool, though a pool can potentially have multiple workloads. Note that the Storage Modeller does not support the
concept of nested pools, such as child pools in IBM Storage Virtualize storage systems.
|
|
Data Space
|
A data space is a logical subdivision of the total storage capacity of a pool. This is a concept introduced by the Storage Modeller to help simplify the modeling of a storage
system. Essentially, a data space is a group of volumes with the same characteristics, within a given pool. The data space (also known as volume) attributes vary by storage system,
depending on the features and capabilities available for the system type and model. For performance modeling, a one-to-one correspondence exists between a workload and a data
space.
|
|
Effective Capacity Used (TiB)
|
In the context of a data space or a workload, the Effective Capacity Used identifies how much of a pool's overall capacity is allocated to that particular data space or workload.
It provides the basic means to subdivide the pool's capacity.
|
|
Volume Allocation
|
In the context of a data space or a workload, the Volume Allocation attribute defines the basic allocation type used for volumes represented by that data space or workload. Volumes
might be “fully allocated” (that is, data reduction is not used for those volumes), or might be “data reduced” if the storage system has the necessary capabilities. Different types
of data reduction exist, for example thin-provisioning, compression, and deduplication. In cases where a storage system is capable of only a specific type of data reduction, the
more specific term (for example, “thin-provisioned”) might be used in favor of the more generic term (for example, “data reduced”).
|
|
Replication
|
For business resilience and disaster recovery, storage system data might be replicated between different storage systems, to provide fault tolerance. Replication relationships are
always uni-directional, from primary to secondary volumes. Individual replication relationships are usually defined for individual volume pairs. However, to simplify storage system
modeling, the Storage Modeller defines replication relationships between individual data space pairs. All volumes represented by a specific data space are assumed to have the same
replication characteristics. There are several types of replication relationships, dependent on the capabilities of the storage system, for example “metro mirror”, “global neap,
and so on.
|
|
Workload Source
|
For performance modeling, a workload might be defined manually by explicitly entering the exact workload characteristics. It might also be automatically generated by specifying a
set of imported data to analyze, from which the Storage Modeller extracts the needed workload characteristics.
|
|
Total I/O Rate (ops/s)
|
The total input/output rate for a given workload or an entire storage system, measured in operations per second. Generally, it is comprised of read and write I/O rates, plus, in
rare cases, additional I/Os that cannot be categorized as read or write (for example, “unmap” I/Os).
|
|
Read I/O Rate (ops/s)
|
The read rate for a given workload or an entire storage system, measured in operations per second. It is included in the total I/O rate.
|
|
Write I/O Rate (ops/s)
|
The write rate for a given workload or an entire storage system, measured in operations per second. It is included in the total I/O rate.
|
|
Read I/O Percentage (%)
|
The break-down of the read vs. write I/O rate for a given workload. For example, a value of 70 indicates that 70% of all operations are reads and 30% are writes.
|
|
Sequential Read I/O Percentage (%)
|
The break-down of sequential vs. random I/O, for the read I/O rate of a given workload. For example a value of 35 indicates that 35% of all read operations are sequential, and 65%
are random. Sequential operations are performed on adjacent logical block addresses (LBAs). In other words, to be sequential, an operation must be performed on an LBA that is
adjacent to the previous operation's LBA.
|
|
Sequential Write I/O Percentage (%)
|
The break-down of sequential vs. random I/O, for the write I/O rate of a given workload. For example, a value of 35 indicates that 35% of all write operations are sequential, and
65% are random. Sequential operations are performed on adjacent logical block addresses (LBAs). In other words, to be sequential, an operation must be performed on an LBA that is
adjacent to the previous operation's LBA.
|
|
Total Synchronous I/O Rate (ops/s)
|
The input/output rate for synchronous operations for a given workload, measured in operations per second. Synchronous operations are performed across IBM DS8000 zHyperLink
connections. They are generally faster than other I/O, but are supported only by DS8000 systems with zHyperLink adapters (and corresponding z/OS host attachments). The synchronous
operations are included in the total operations. They are part of the total I/O rate. Hence, a workload's total synchronous I/O rate must not exceed the workload's total I/O rate.
|
|
Read Synchronous I/O Rate (ops/s)
|
The synchronous read rate for a given workload, measured in operations per second. It is included in the total synchronous I/O rate.
|
|
Write Synchronous I/O Rate (ops/s)
|
The synchronous write rate for a given workload, measured in operations per second. It is included in the total synchronous I/O rate.
|
|
Total zHPF I/O Rate (ops/s)
|
The input/output rate for System Z “high performance FICON” operations for a given workload, measured in operations per second. zHPF operations are performed across IBM DS8000
FICON connection with zHPF capability. They are generally faster than other FICON I/O (with the exception of synchronous I/O), but only DS8000 systems with zHPF capable adapters
(and corresponding System Z host attachments) support them. The zHPF operations are included in the total operations. They are part of the total I/O rate. Hence, a workload's total
zHPF I/O rate must not exceed the workload's total I/O rate.
|
|
Read zHPF I/O Rate (ops/s)
|
The zHPF read rate for a given workload, measured in operations per second. It is included in the total zHPF I/O rate.
|
|
Write zHPF I/O Rate (ops/s)
|
The zHPF write rate for a given workload, measured in operations per second. It is included in the total zHPF I/O rate.
|
|
Overall Transfer Size (KiB/op)
|
The average transfer size of all input/output operations for a given workload, measured in KiB (1024 bytes) per operation. Generally, this average is computed from the read and
write transfer sizes, weighted by read and write I/O rates.
|
|
Read Transfer Size (KiB/op)
|
The average transfer size of all read operations for a given workload, measured in KiB (1024 bytes) per operation. This value is incorporated into the overall transfer size.
|
|
Write Transfer Size (KiB/op)
|
The average transfer size of all write operations for a given workload, measured in KiB (1024 bytes) per operation. This value is incorporated into the overall transfer size.
|
|
Total Data Rate (MiB/s)
|
The total throughput of all input/output operations for a given workload, measured in MiB (1,048,576 bytes) per second. It is comprised of read and write data rates.
|
|
Read Data Rate (MiB/s)
|
The total throughput of all read operations for a given workload, measured in MiB (1,048,576 bytes) per second. It is included in the total data rate.
|
|
Write Data Rate (MiB/s)
|
The total throughput of all write operations for a given workload, measured in MiB (1,048,576 bytes) per second. It is included in the total data rate.
|
|
Total Cache Read Hits (%)
|
The percentage of all read operations that find their data in cache, for a given workload.
Caution: Cache hit percentages depend, to a great degree, on the amount of cache available in the specific storage system that is modeled. This means that
values are not directly transferable from one storage system to another, without adjusting for cache size differences.
|
|
Cache Random Read Hits (%)
|
The percentage of all I/O operations that are random read operations which find their data in cache, for a given workload. It is equivalent to the total cache read hits percentage
expressed as a percentage of all I/Os (rather than just read I/Os), minus the cache sequential read hits percentage.
|
|
Cache Sequential Read Hits (%)
|
The percentage of all I/O operations that are sequential read operations that find their data in cache, for a given workload. It is equivalent to the cache sequential read miss
rate expressed as a percentage of all I/Os.
|
|
Total Write Cache Efficiency (%)
|
The percentage of all write cache hits that occur on data that is marked as modified in the cache, for a given workload. This value represents how effectively write operations are
coalesced before the data is written to disk. This value incorporates both the random and sequential write cache efficiency, weighted by the sequential and random write I/O
percentages.
|
|
Random Write Cache Efficiency (%)
|
The percentage of all random write-cache hits that occur on data that is marked as modified in the cache, for a given workload. This value represents how effectively random write
operations are coalesced before the data is written to disk. This value is included in the total write cache efficiency percentage.
|
|
Sequential Write Cache Efficiency (%)
|
The percentage of all sequential write cache hits that occur on data that is marked as modified in the cache, for a given workload. This value represents how effectively sequential
write operations are coalesced before the data is written to disk. This value is included in the total write-cache efficiency percentage.
|
|
Record Cache Read Miss Rate (ops/s)
|
The rate of all System Z record-mode read operations that do not find their data in cache, measured in operations per second.
|
|
Cache Sequential Read Miss Rate (trks/s)
|
The rate of all sequential read operations that do not find their data in cache, measured in tracks per second. Essentially, this is the number of tracks transferred per second
from disk to cache for sequential read operations, i.e. the pre-staging rate.
|
|
Cache Random Write Miss Rate (trks/s)
|
The rate of all random write operations that do not find their data in cache, measured in tracks per second. Essentially, this is the number of tracks transferred per second from
cache to disk for random write operations (that is, the random write destage rate).
|
|
Cache Sequential Write Miss Rate (trks/s)
|
The rate of all sequential write operations that do not find their data in cache, measured in tracks per second. Essentially, this is the number of tracks transferred per second
from cache to disk, for sequential write operations (that is, the sequential write destage rate).
|
|
Easy Tier Skew Factor
|
The skew factor is a measure of the uniformity of the distribution of I/O across the logical blocks in a storage pool. A skew factor of 1.0 describes a perfectly uniform
distribution, where each of the pool's logical blocks is targeted by approximately the same number of I/O operations. A higher skew factor means that a smaller number of logical
blocks is targeted by a larger number of operations. For example, a skew factor of 7.0 means that 25% of the pool's logical blocks get 80% of all operations, and a skew factor of
24.0 means that 10% of all logical blocks get around 90% of all operations. The higher the skew factor, the less uniform the distribution of I/O, and the better suited the pool
is for Easy Tier operation.
Easy Tier works on pools that contain a mix of drive technologies, and relies on placing the most used (the “hottest”) blocks on the fastest drives, and the least used (the
“coldest”) blocks on the slowest drives, to maximize the overall performance of the pool. The greater the distinction is between hot and cold blocks, the more effective is the
Easy Tier performance optimization.
|