 |
|
|
|
|
|
|
|
When vmstat indicates I/O wait times
in case of an I/O bound workload, the latency of the storage
subsystem limits the transactional throughput of the database.
Using several storage servers in parallel can improve this
limitation.
The following chart shows an example
that uses storage servers of two different types named Type 1 and
Type 2. Type 2's capacity is similar to the capacity of two servers
of Type 1.
The usage of two Type 1 storage
servers increases the throughput about 50% in comparison to a
single Type 1 storage server.
Using all three storage servers
together again increases the throughput about 50%. In all cases for
the Linux point of view the database files are identical, only the
disks belong to different storage servers.
- IBM eServer zSeries 900 (2064-216)
- LPAR with 8 CPUs, 2GB memory
- Novell/SUSE SLES 8 + SP2 for IBM zSeries (64-bit),
kernel 2.4.19
- Informix 9.4.0 FC3 (64-bit)
- Benchmark Informix OLTP
- Database data Logical Volume (LV): 8 CHPIDs, 8 host
adapters, FCP/SCSI or FICON/ECKD disks in 8 ranks,
ext2 filesystem
Conclusion: You can achieve a significantly higher
I/O bandwidth if you distribute your database disks
over several disk subsystems.
|
|
|
|
|
|
