OpenSIUC Articles Department of Electrical and ComputerEngineering 2008 Enabling Location-Based Services in Data Centers

In this article, we explore services and capabilities that can be enabled by the localization of various assets in a data center or IT environment. We also describe the underlying location estimation method and the protocol to enable localization. Finally, we present a management framework for these services and present a few case studies to assess benefits of location-based services in data centers. Enabling Location-Based Services in Data Centers KANT LAYOUT 11/5/08 3:39 PM Page 20 Authorized licensed use limited to: Southern Illinois University Carbondale. Downloaded on May 30, 2009 at 15:28 from IEEE Xplore. Restrictions apply. IEEE Network • November/December 2008 21 an RFID solution is neither cost effective nor can it achieve the desired localization accuracy. Localization is a very well-studied problem in wireless networks; however, our interest is in only those technologies that are accurate enough to locate individual racks/chassis and (preferably) individual servers. Note that the localization of 1U servers requires accuracies of the order of one inch. In the following, we survey some localization technologies and address their applicability to data centers. WLAN-based localization is extensively explored in the literature [4] and can be implemented easily in software. Unfortunately, even with specialized techniques such as a multipath decomposition method [4], the root mean square error (RSME) in the best line-of-sight (LoS) case is only 1.1 meters. Ultrasonic or surface acoustic wave (SAW) systems perform localization based on time of flight (ToF) of sound waves. Because of the very low speed of sound, SAW systems can measure distance with an accuracy of a few centimeters. Unfortunately, SAW systems require substantial infrastructure and uninterrupted sound channels between emitter and receivers. In [2, 3], we explored a WUSB-based localization solution that assumes that each server comes fitted with a WUSB radio (as a replacement for, or in addition to, the wired USB interface) that has requisite time of arrival (ToA)-based measurement capabilities. This can provide an effective and inexpensive localization solution. WUSB Standardization and Platform Issues The IEEE standards group on personal area networks (PANs) is actively working on UWB-based communications under the WiMedia alliance and 802.15.4 task group. WUSB is a middleware layer that runs atop WiMedia medium access control (MAC). 802.15.4a focuses on low data rate (LDR) applications (≤ 0.25 Mb/s), and is set to serve the specific requirements of industrial, residential, and medical applications. The design of 802.15.4a specifically addresses localization capability and is ideally suited for LBS applications. Our suboptimal choice of WUSB/WiMedia is motivated by practical considerations: as stated above, we expect WUSB to be ubiquitous; therefore, using WiMedia does not require additional expense or complexity for data center owners. Of course, everything about the proposed techniques (with the exception of timing) applies to 802.15.4a as well. WUSB uses the MAC protocol based on the WiMedia standard. It is a domain-dependent MAC with a master-slave architecture involving a Piconet controller (PNC) and up to 255 terminals (slaves). The PNC maintains global timing using a super frame (SF) structure. The SF consists of 256 slots, and each slot has a duration of 256 microseconds. An SF consists of a beacon period, contention-access period, and contentionfree period. The beacon period is used for PNC to terminal broadcasts; a contention-access period is used by the terminals to communicate with others or to ask the PNC for reservedchannel time; and a contention-free period is dedicated for individual transmissions over agreed upon time slots. Server localization is often a crucial functionality when the server is not operational (e.g., due to replacement, repair, or bypass). Consequently, the localization driver is best implemented in the baseboard management controller (BMC) of the server rather than in the OS of the main processor. BMC is the main controller that will stay operational, as long as the server is plugged in and provides for intelligent platform management [5]. However, providing BMC control over WUSB in a post-boot environment is a challenge that is not addressed here. Location Estimation Methods Localization involves determining the position of an unknown node in a twoor three-dimensional space using range estimates from a few reference nodes, (i.e., nodes with known locations) to an unknown node. The range estimate can be obtained using received signal strength (RSSI), a ToA technique, an angle of arrival (AoA) technique, or a hybrid method that is a combination of any of these methods. Here, we focus on the most widely used ToA method for UWB ranging. The ToA technique determines the distance by estimating the propagation delay between the transmitter and receiver. Then, the position of an unknown node is identified using the traditional methods such as the intersection of circles or the intersection of hyperbolas using the time difference of arrival between the two ToAs [6]. However, due to errors in range measurements, a statistical estimation technique such as maximum likelihood estimation (MLE) is required. MLE estimates distributional parameters by maximizing the probability that the measurements came from the assumed distribution. Because the server positions can take only a small number of discrete positions in a rack, the MLE problem can be transformed into a simpler maximum likelihood identification (MLI) problem [3]. MLI exploits the geometry of racks to accurately identify the position of the unknown server. Figure 2 shows the rack configuration and an associated coordinate system (x, y, z), where x is the row offset, y is the rack offset within a row, and z is the server height in a rack. Consider rack(0,0) with N plugged-in servers. For determining the location of unknown server u, MLI uses three reference nodes, of which the first two are in rack(0,0) and the third one in rack(0,1). Each reference node i (where i ∈ 1, 2, 3) measures the distance to an unknown node u as riu using ToA. We assume that a range estimate riu is distributed as Gaussian with zero bias (that is, the expected value of the estimate equals true distance) and variance of σ2 = N0/2. The distance between each reference node and N-2 possible positions in the rack is known. Given the three range estimates and N-2 possible distances from each of the reference nodes, N-2 likelihood functions (LFs) are formed. Out of N-2 LFs, the minimum-valued LF identifies the position of an unknown server. In [3], it is shown that the performance of the MLI method far exceeds the performance of the traditional methods. Localization Protocol Asset localization in data centers involves the following two distinct phases: • Cold-start phase that localizes all servers starting with a few reference servers with known locations n Figure 1. Snapshot of row of a typical data center. KANT LAYOUT 11/5/08 3:39 PM Page 21 Authorized licensed use limited to: Southern Illinois University Carbondale. Downloaded on May 30, 2009 at 15:28 from IEEE Xplore. Restrictions apply. IEEE Network • November/December 2008 22 • Steady-state phase that tracks individual asset movements subsequently The steady-state phase is relatively easy to handle and is not described here due to space constraints. The cold-start phase starts with one of the known servers in the servers that are hard coded as PNC and all others in the listening mode. The role of the PNC is to form the Piconet with the servers from the current rack and a few servers from the adjacent and the opposite rack to enable rack-to-rack localization. One complication in cold-start localization is the avoidance of servers in racks that we are currently not interested in localizing. This, in turn, requires “macro-localization,” that is, the determination of which rack the responding servers belong to, so that we can suppress the undesirable ones. This is handled by a combination of careful power control and by exploiting the geometry of the racks. Generally the localization proceeds row by row as explained below. Row 0 Localization — We start with three known servers as shown in Fig. 2. During rack(0,0) localization, all the unknown servers in rack(0,0) and at least one server in the adjacent rack(0,1) and two servers in the opposite rack(1,0) are localized to enable localization in the subsequent racks as shown by red and green/black arrows in Fig. 2. (To avoid clutter, not all arrows are shown.) After the current rack localization is complete, the PNC in the current rack performs hand off to one of the localized servers (new PNC) in the rack(0,1). Thus, localization continues one rack at a time along with a few localizations in the adjacent and opposite rack until all servers in the last rack of row 0 are localized. After the last rack localization, PNC in the last rack updates all the servers with the position of their neighbors and hands off to the selected PNC in the last but one rack in row 0. This hand off in the reverse direction continues until the rack(0,0) is reached. Now PNC in rack(0,0) is ready to hand off to the suitable known server in the rack(1,0) (odd numbered row). Row 1 Localization — At the beginning of the row 1 localization, all the servers in row 0 are localized, and the PNC in rack(0,0) selects a known server as a new PNC in rack(1,0). In the beginning of row 1 localization, each rack in row1 has at least two known servers. But, there are no known servers in row 2. Also, given the alternating rows of front and back facing servers, communication across the back aisles is very challenging due to the heavily metallic nature of the racks as shown in Fig. 2. Therefore, only the racks located at the edge of the one row can communicate with the racks located at the edges of the next rows. During rack(1,0) locali