HP-UX Encrypted Volume and File System Performance and Tuning

8
Testing Methodology
Many types of transactions and protocols benefit from standardized specifications for testing bodies
and test suites. This is not true for data encryption. Considering the lack of specifications, the
clearest method to measure and communicate the performance effects of data encryption is to
measure identical tests while reading and writing clear data versus encrypted data, then compare the
results. For these tests, the HP-UX JFS volume configuration granularity of EVFS was utilized to
configure identically sized data volumes on the same storage device and the same HP-UX system, so
there is no variation in equipment or system configuration. All variables are identical except that
EVFS throughput is measured while traversing the EVFS Pseudo-Driver, where clear I/O is measured
using the JFS file system stack.
Measures
Performance for EVFS is measured in throughput by megabytes per second, versus the system
resources required to achieve that level of throughput. Memory utilization has been proven to be
minimal and does not significantly scale with throughput, so memory utilization will not be reported in
the test case graphs. CPU utilization does scale with throughput, and is displayed in relation to
throughput.
Each test case is run on a clear-data volume to establish a baseline throughput level. The clear data
baseline is then optimized by testing for effective kernel tuning and file system tuning. This baseline is
then used as the EVFS baseline, which is then further optimized by testing for effective kernel tuning
and file system tuning for EVFS.
All tests are executed three times, and the results are averaged and reported using the graphed
metrics.
Benchmark
To drive these tests, the defacto standard IOZone application was used. IOZone is widely used as a
quick and effective way to drive I/O throughput to infinitely configurable levels, and to accurately
measure the throughput on a consistent scale.
IOZone is obviously I/O intensive, and is certainly not representative of a typical customer
application. While a typical application may use CPU to process data before reading or writing it,
IOZone uses comparatively less CPU and emphasizes the volume of read and write data. Thus,
IOZone is probably a “worst-case scenario” for EVFS. These tests should not be construed as
indicators of how an actual application would perform with EVFS. These are simply
extreme loads that provide a contrast between clear I/O throughput and encrypted I/O throughput.
Postmark - another benchmark utility - was used to provide a more application-like I/O throughput
representation. Postmark was developed to emulate a mail server file system environment, where
many small files are constantly created, deleted, and modified. The system load emulation derived
from Postmark is very different from the massive read and write operations used by IOZone. Postmark
also measures Transactions Per-Second, which are displayed in the Results chapter.
System Configuration
The system used for these tests was an HP Integrity Server rx7620: