Filter
Top Products
Chapter 12. Performance considerations 271
the I/O rate and highly sequential read operations for the MB/sec numbers. They also vary
depending on the server type used.
The 2.8 GB/sec sequential read figure—and even significantly more—is achievable with a
properly configured DS8300.
A properly configured DS8100 can reach read hit I/O rates with numbers in the 6 digits. The
DS8300 does more than double that of the DS8100. These are rough estimates based on the
technology and architectural possibilities, and are presented just to help you start imagining
what might be possible. More precise figures cannot be stated without access to benchmark
results or your own benchmark experience.
What we can expect to see is a significant improvement in response time and throughput with
cache-hostile workloads due to the very fast Fibre Channel switched disk subsystems. A new
caching algorithm, SARC, is used to guarantee a more efficient cache management than with
the old LRU approach. This, in turn, will also positively contribute to better response times for
all workloads which show pure locality of reference with the LRU approach.
12.5.3 Appropriate DS8000 size in z/OS environments
The potential of the architecture, its implementation and utilized technology allow for some
projections at this point (though without having the hard figures at hand). Rules of thumb have
the potential to be proven wrong. Therefore, you see here some recommendations on sizing
which are rather conservative.
A fully configured ESS 800 Turbo with CKD volumes only and 16 FICON channels is good for
the following:
Over 30,000 I/Os per second
More than 500 MB/sec aggregated sequential read throughput
About 350 MB/sec sequential write throughput mirrored in cache
Without discrete DS8000 benchmark figures, a sizing approach to follow could be to propose
how many ESS 800s might be consolidated into a DS8000 model. From that you can derive
the number of ESS 750s, ESS F20s, and ESS E20s which can collapse into a DS8000. The
older ESS models have a known relationship to the ESS 800.
Further considerations are, for example, the connection technology used, like ESCON,
FICON, or FICON Express channels, and the number of channels.
Generally speaking, a properly configured DS8100 has the potential to provide the same or
better numbers than two ESS 800s. Since the ESS 800 has the performance capabilities of
two ESS F20s, a properly configured DS8100 can replace four ESS F20s. As the DS8000
series scales linearly, a well configured DS8300 has the potential to have the same or better
numbers concerning I/O rates, sequential bandwidth, and response time than two DS8100 or
four ESS 800s. Since the ESS 800 has roughly the performance potential of two ESS F20s, a
corresponding number of ESS F20s can be consolidated. This applies also to the ESS 750,
which has a similar performance behavior to that of an ESS F20.
Based on customer workload data, an IBM internal modelling tool can project how many ESS
800s to configure and then consolidate the ESS 800s to respective DS8000 models.
Processor memory size considerations for z/OS environments
Processor memory or cache in the DS8000 contributes to very high I/O rates and helps to
minimize I/O response time.