![](bg1.png)
Data Location Optimization Method to Improve Tiered
Storage Performance
Shinichi Hayashi
1
and Norihisa Komoda
2
1
Yokohama Research Laboratory, Hitachi, Ltd., Kanagawa, Japan
2
Graduate School of Information Science and Technologies, Osaka University, Osaka, Japan
Keywords: Data Location Optimization, Tired Storage, Dynamic Tier Control, I/O Performance, SSD.
Abstract: We propose a new method for tiered storage and evaluate the characteristics of the method. In this method,
a fast tier is divided into two areas, and the data in each area is managed on the basis of two input/output
(I/O) measurement periods. When I/Os to these areas increase, the proposed method allocates frequently
accessed areas to a higher tier in advance, therefore, improving the total system I/O performance. When
frequently accessed areas rarely move, the proposed method is the most effective and improves the total
system I/O performance by up to 30.4%.
1 INTRODUCTION
With recent improvements in information
technology, the amount of data retained by
companies has increased exponentially. The capacity
of hard disk drives (HDDs) has continued to
increase; however, the performance of these devices
has not improved significantly. Therefore, HDDs
can potentially become bottlenecks. Solid state
drives (SSDs), which are much faster than HDDs,
are currently attracting attention. When HDDs are
bottlenecks, replacing HDDs with SSDs could
potentially increase performance. However, SSDs
are generally more expensive and have less storage
capacity. Therefore, budget and capacity constraints
limit a company’s ability to replace all HDDs with
SSDs.
In General, input and output (I/O) activities have
a locality, and the number of I/Os for each storage
area is different. Therefore, if frequently accessed
areas are stored on a SSD and rarely accessed areas
are stored on a HDD, I/O performance will increase
and system cost will be reduced. A frequently
accessed area is denoted as a hot area and a rarely
accessed area is denoted as a cold area.
We define media in different levels of
performance as storage tiers and define storage that
leverages several tiers as a tiered storage. Several
reports have investigated tiered storage, and several
companies provide tiered storage (Hitachi, 2012)
(Schmidt et al., 2012) (EMC, 2012) (Hewlett-
Packard, 2012). Tiered storage moves data to an
appropriate tier on the basis of the number of I/Os,
as described above. In this paper, this function is
denoted as “Dynamic Tier Control.” In Dynamic
Tier Control, the storage measures the number of
I/Os for each area for a certain period, which is
denoted as the “I/O measurement period.” On the
basis of this measurement, the stored data is moved
to an appropriate tier. I/O measurement period also
indicates the frequency of data movement between
tiers.
To increase system performance, hot areas
should be allocated to a faster tier. This could be
accomplished by shortening the I/O measurement
period, because areas that become hotter are moved
to a faster tier immediately. However, this would
result in frequent data movements between tiers and
would have a negative impact on regular I/O
between servers and storage. On the other hand, if
the I/O measurement period was longer, data
movements between tiers would be reduced.
However, in this case, cold areas would remain
allocated to the faster tier. As a result, system I/O
performance would decrease.
In this paper, to address the above problem, we
propose a new fast tier allocation method on the
basis of short and long I/O measurement periods.
Since the effects of our proposed method will differ
depending on application I/O characteristics, I/O
measurement periods, and the rates of capacity of
112
Hayashi S. and Komoda N..
Data Location Optimization Method to Improve Tiered Storage Performance.
DOI: 10.5220/0004415901120119
In Proceedings of the 15th International Conference on Enterprise Information Systems (ICEIS-2013), pages 112-119
ISBN: 978-989-8565-59-4
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)