Hitachi STORAGE 9900 - Manual (Page 41)

31
Unmatched Performance
and Scalability
Truly Scalable Performance
Based on the internal switched Hi-Star ™ architecture, the Hitachi Freedom Storage ™
Lightning 9900 ™ V Series shows unmatched performance scalability. This chapter discusses
the ability to dramatically add more online transaction processing (OLTP) users and mixed
decision-support and OLTP workloads without a user response time decrease. The mor e
capacity and connectivity added to a Lightning 9900 V Series system, the more total the
bandwidth increases. As Table 4 shows, total aggregate bandwidth is ten times that of
competition.
When new application workloads are added to a Lightning 9980V ™ system with an
existing workload, neither the new nor the existing workloads suffer performance
degradations from the sharing of internal resources and bandwidth. The following series of
tests proves this point. By measuring user-response time while workloads are added to the
system, we can illustrate this phenomenon.
I n t e rnal Switching Overcomes the Limitations of Shared-bus Arc h i t e c t u re s
It can be seen from Figure 24 that very little user-response time reduction occurs in the
OLTP performance of the Lightning 9980V system. This is because the Lightning 9900 V
Series system, with its Hi-Star architecture, essentially has its own internally routed path for
each workload. This means that the main reason for performance decline of a competitive
shared-bus architecture, i.e. bus contention, does not occur with Lightning 9900 V Series
systems. Although there may be some contention in cache, many simultaneous operations
concurrently happen at the cache, so it far exceeds the demands of most applications.
T h e re Is NO Match for the Performance of the Lightning 9980V System
As shown in Figure 23, during the first 30 minutes of the test 1800 simulated users
were connected to the first OLTP server, and the transactions per minute (TPM) rate and
user-response time reached steady state at about 2200TPM and 3.0 seconds respectively.
OLTP user response time is the delay time between when a simulated user issues a
transaction request and when the request responds with a commit. After 30 minutes, a
second OLTP server was connected with an additional 1800 simulated users, bringing the
total number of users to 3600.
After one hour, the first decision support system (DSS) workload was started. The DSS
workload began a series of full table scans, counting records from a 1.2 billion row table.
The transfer rate for the DSS system was 380MB/sec or 3.4 million rows per second. We
8
Bottleneck-free
performance with the
Lightning 9900 V Series.
Table 4 – Lightning
9980V system bandwidth
is ten times that of
competition.
The more capacity and
connectivity added to a
Lightning 9900 V Series
system, the more total
internal bandwidth
increases.
Model Control Data Aggregate User-useable
Type Bandwidth Bandwidth Bandwidth Throughput
Lightning 9970V ™ system
2.6GB/sec 5.3GB/sec 7.9GB/sec 5.3GB/sec
Single Cabinet
Lightning 9980V ™ system
5.3GB/sec 10.6GB/sec 15.9GB/sec 10.6GB/sec
Multi-Cabinet
Competition 1.6GB/sec 1.6GB/sec 1.6GB/sec 1.6GB/sec