HP SmartCache technology

Technical white paper | HP SmartCache
10
Furthermore, without HP SmartCache, we can see that
the average read latency per request was almost 28
milliseconds and the average write latency was
practically non-existent at 0.13 milliseconds (Figure 8).
This workload highlights the benefit of the FBWC of
the Smart Array controller. The FBWC of the Smart
Array controller completes the write operation by
saving the data within the FBWC memory, which is
much quicker than writing the data out to the hard disk
drives.
Once data is saved in the FBWC, the Smart Array
controller ensures that the data is written to the hard
disk drives. The primary function of the FBWC is to
protect the data in the FBWC in case of unplanned
power loss to the server. From an application
workload perspective, the write load of the application
o the Smart Array controller is not high enough nor
sustained long enough to fill the capacity of the FBWC
so the write IO response time is excellent.
Figure 9 also shows that read latency drops from almost 28 to 6.23 milliseconds per request when HP SmartCache is
enabled. This behavior is expected as more reads are satisfied by the accelerator device or SSD, which demonstrates the
value of HP SmartCache. What might not be expected is the increase in write latency when HP SmartCache is enabled, which
will be explained shortly.
Looking at the distribution of latencies per transaction for 8K read traffic (Figure 9) shows the benefits of HP SmartCache.
With HP SmartCache enabled, 95% of the read requests completed in under10 milliseconds. Without HP SmartCache, the
distribution of latencies is spread out over the spectrum of response times as there is more dependency on the rotation
access time of the hard disk drives. This rotational latency results in a near
28-millisecond average per read request.
Figure 9. 8K Read Latency Distribution
Write latency jumped significantly from an average of 0.13 milliseconds without HP SmartCache to 89.35 milliseconds with
HP SmartCache (Figure 10). Without HP SmartCache, the write requests were completed with low latency by the FBWC. With
HP SmartCache, the increase in read requests serviced by the faster SSD results in a workload shift on the HDDs from
mostly reads to mostly writes.
Note also that using HP SmartCache results in the volume of writes from the application exceeding the FBWC capacity.
When the FBWC memory is full, the write latencies increase since write operations cannot complete until data is written to
the HDDs. This explains the increase in write latencies when using HP SmartCache. However, the benefit of substantially
increased read performance far outweighs the increase in write latency, resulting in better overall performance to the
application.
Milliseconds
8K read, no cache
8K read, SmartCache
Amount of reads
Figure 8.
Average latency per I/O.
No cache
SmartCache
Reads Writes
Latency in milliseconds