175
Copyright@2004~2014 Qsan Technology, Inc. All Rights Reserved.
Qsan Document – User Manual
application to get the best performance. If the above three applications are not suitable, the last
item is customization which you may set the configurations by yourself.
I/O Type
Block Size
(Sectors)
Sub-block Size
(Sectors)
Populate-on-
Read Threshold
Populate-on-
Wrote Threshold
Database
1MB (2,048)
8KB (16)
2
0
File System
2MB (4,096)
16KB (32)
2
2
Web Service
4MB (8,192)
64KB (128)
2
0
Customization
1MB/2MB/4MB
8KB/16KB/64KB
≥ 0
≥ 0
The block size affects the cache use and the warm up time. The cache use shows how much of the
allocated cache actually holds the user data. And the warm up time is the process of how long to
fill the cache. You can image that the highest cache use is obtained when all of the frequently
reread data is located very close to other data that is frequently reread. Using a larger cache block
size of I/O type is more useful to performance than a smaller one. Conversely, when frequently
reread data is located far from other data that is frequently reread, the lowest cache use is
obtained. In this case, the lowest cache block size of I/O type allows the most user data to be
cached.
The sub-block size affects the cache warm up time, too. A larger sub-block size causes cache to fill
more quickly than a smaller one, but it can also affect the response time of host I/O. Also occupy
the system resource, such as CPU utilization, memory bandwidth, or channel utilization. A very
high locality of reference can be more useful from a larger sub-block size than from a smaller one,
especially if those blocks that are reread frequently reside in the same sub-block. This occurs when
one I/O causes the sub-block to be populated and another I/O in the same sub-block gets a cache
hit.
These are trade off depend on the applications. Users may set them by experience to get the best
performance. Here we provide a formula which can calculate the estimate warm up time.
We define that
T: Warm up time; seconds required.
I: Best random IOPS of HDD.
S: I/O Size.
D: Number of HDDs.
C: Total SSD caching capacity.
P: Populate-on-read or Populate-on-write threshold.
We assume that random read/write from HDD to achieve the capacity of SSD should be
