SPC-1 Like Workload

SPC-1 is a synthetic, but sophisticated and fairly realistic, performance measurement workload for storage subsystems used in business critical applications. The benchmark simulates real world environments as seen by on-line, non-volatile storage in a typical server class computer system. SPC-1 measures the performance of a storage subsystem by presenting to it a set of I/O operations that are typical for business critical applications like OLTP systems, database systems and mail server applications. For extensive details on SPC-1, please see: [23,24]. We used SPC-1 Like that is an earlier prototype implementation of SPC-1 benchmark by Bruce McNutt who was one of the chief architects of the official SPC-1 benchmark.

The SPC-1 Like workload synthesizes a community of users targeting I/Os to the storage that is logically organized in the form of three Application Storage Units (ASU). ASU-1 represents a ``Data Store'', ASU-2 represents a ``User Store'', and ASU-3 represents a ``Log/Sequential Write''. Of the total amount of available back-end storage, 45% is assigned to ASU-1, 45% is assigned to ASU-2, and remaining 10% is assigned to ASU-3 as per SPC-1 specifications. We shall furnish more details on sizes of various ASUs in Section IV-D.

The SPC-1 Like workload is specified in units of Business Scaling Units (BSU). One BSU corresponds to a community of users who collectively generate up to 50 IOPS. The overall composition of a BSU, and, that of SPC-1 Like, is specified by the following simple matrix, where all numbers are in percentages:

	Read	Write	All
Random	29	32	61
Sequential	11	28	39
All	40	60	100

The workload is scaled by using more BSUs that in effect increases the number of users being simulated.

In this paper, due to the commercial nature of the system involved, we will not use IOPS, but rather use scaled IOPS which are obtained by multiplying the true IOPS by a constant (a non-integer rational number) that is not revealed in the paper.

We shall use two different load schedules for SPC-1 Like in Table I.

Table I: The structure of the SPC-1 Like benchmark on two different loads. For both loads, the warm-up phase is run for $120$ minutes followed by $6$ measurement phases of $30$ minutes each. The purpose of the warm-up phase is to fill-up the cache and to brings it to a steady-state. Observe that in the measurement phase, we gradually decrease the load in proportion $100\%$, $97.5\%$, $95\%$, $80\%$, $50\%$, and $10\%$, where $100\%$, represents the highest load. This allows studying the behavior of the storage controller under a wide range of load conditions.

	Warm-up	Measurement
	Phase	Phase
Time (in minutes)	120	30	30	30	30	30	30
Load (percentages)	100	100	97.5	95	80	50	10
High Load (scaled IOPS)	25000	25000	23750	22500	20000	12500	2500
Low Load (scaled IOPS)	11364	11364	10795	10227	9091	5682	1136