Benchmarking I/O with Oracle ORION is an important part of planning, baselining, and performance-tuning Oracle environments. I’ve previously provided ORION results for the hi1.4xlarge SSD-backed instance class, and based on some recent work, I wanted to provide an update for the newer hs1.8xlarge instance class. Below you’ll find hs1.8xlarge Oracle ORION benchmark results with the following configuration:
- Vanilla RHEL 6.4 (ami-a25415cb)
- No blkdev, ioctl, or sysctl tuning
- All 24 attached drives in orion.lun
- No ephemeral warmup for hs1.8xlarge
- CONCAT VLUN, no striping -simulate
ORION Run #1, hs1.8xlarge: -run oltp -write 20
- Maximum Small IOPS=16211 @ Small=336 and Large=0
- Minimum Small Latency=2.71 @ Small=24 and Large=0
ORION Run #2, hs1.8xlarge: -run dss
- Maximum Large MBPS=2875.10 @ Small=0 and Large=336
In summary, our tests imply that:
- For 80% read, 20% write OLTP workloads, an hs1.8xlarge is capable of sustaining 16K IOPS per second with 14 (336/24 devices) concurrent small (8K) operations per device.
- For 80% read, 20% write OLTP workloads, an hs1.8xlarge can provide sub-5ms operation latencies for single operations per device.
- For read-only DSS/data warehouse workloads, an hs1.8xlarge can push approximately 2.8GB/sec at 14 (336/24) concurrent large (1MB) operations per device.
For application architectures where high-availability or performance can compensate for the durability of ephemeral storage, hs1.8xlarge instances provide incredibly compelling performance on an on-demand or reserved basis.
Will you share how you configured the SSD drives using CONCAT VLUN, no striping -simulate ?
I am only getting 6k IOPS when I use lvm2 with concating(no striping) on my hs1.8xlarge instance.