Hearing loss is a major health issue in the work environment. Exposing humans to excessive sound levels or frequencies can adversely impact the ability of workers to communicate or hear. Noisy environments inside data centers present a unique occupational safety exposure to staff and operators which spend a considerable amount of time in the rooms to perform daily tasks. The specific sound frequency emitted by servers may also have a negative impact on worker performance and well-being which has not been analyzed in the past as this is an emerging technology. Specific opportunities may exist by analyzing the characteristics of the sound signal produced from servers in the server rack partially cancel the sound waves through similarity, time-delay and the correlation for the produced signals.
Occupational Noise. The assessment of hearing loss is performed by audiologists and is a function
of the shifting of the hearing threshold as defined by audiometry. The degradation of hearing associated
with excessive noise is irreversible
Noise Inside Server Rooms. Network engineering technicians and other workers access server rooms to install, fix and configure devices on a regular basis. Noisy environments inside data centers represent an occupational exposure tostaff, especially those which spend a considerable amount of time in the rooms to perform daily tasks. This noise may also cause discomfort for operators even if the noise inside the rooms is below the OSHA mandated threshold.
Sound is measured in Decibels (dBA) (A-Scale more closely represents hearing exposures for
humans) which is a unit of sound pressure in Pascal (Pa)
For example, it is theorized that working in a data center noise may cause tinnitus, which is an
unpleasant ringing in a person’s ears) due to the frequency of the sound and not the overall sound power.
Alternative Solutions for Server Room Noise. There are several approaches for mitigating the impact
of server room noise on the employees evaluated. It is important to note that the sound pressure values
did not exceed federal mandates requirement. However, this requirement is based upon the OSHA
regulations which are over 40 years old (OSHA)
The issue here is the frequency of the sound and not the sound pressure and addressing employee’s
concerns where science and epidemiology do not appear to be up-to-date on the impact of this kind of
technology. The most common technique to reduce the occupational noise exposure is by reducing the
sound volume by administrative procedures (i.e. reducing time of the operators in the server room), or
through personal protective equipment like earplugs. The construction of the earplug with soft silicon
or a wax pillow allows the user to insert this device into the ear canal to reduce the pressure of the noise
on the inner part of the ear. The use of these plugs is effective in filtering the noise if the employee uses
them correctly, but it impacts their ability to communicate with other workers inside the server room.
Workers may stay for a prolonged duration inside the data center to conduct maintenance work and
communication within parties especially during times of crisis is imperative. New technology that
allows employees to communicate is evolving, but these solutions are expensive and require that the
company purchase units for each employee that are over $1000 a unit and require potential fitting the
employee’s ear
Earmuffs are another solution for noise inside server rooms. Earmuffs cover the pinna (the external
part of the ear) to reduce the energy of the noise that reach the inner ear of the human
OSHA requires the use of hearing protection such as earplugs and earmuffs only when there are no
feasible cost-effective engineering or administrative solutions for the noise mitigation. OSHA will also
recommend engineering over administrative controls since administrative controls require changes in
the schedule or workplace to reduce or eliminate exposure of workers to source of noise
Currently, many Human Factors Engineers are exploring the efficacy of active noise control that
may serve for noise reduction in a data center. Active noise control methods have received some
attention as they do not require any noise absorbing material
There are different algorithms that can be applied for the controller such as the least mean square (LMS) filter, the recursive least square (RLS) algorithm and the filtered reference least mean square (FxLMS). Active Noise Control methods are complicated and expensive to deploy in large server rooms, and therefore may not be cost-effective to reduce noise.
Data Collection Tools. The noise volumes in the server room were collected twice in different manners. Firstly, the data were captured utilizing the Sound Pro Sound Level meter, which is manufactured by Quest Company (Model 2900/Type 2) and referenced in Fig. 1.
Sound level data were collected in 12 locations inside an actual server room that was evaluated (Fig. 2). It is worth to mention that size of the room 20 ft width, 15 ft length and 10 ft height with only two server’s racks. The layout of the room and servers is depicted in the figure along with the twelve respective locations (A-13, M-15, W-13, I-19, M-18, R-19, A28, M-21, W-28, D6-2, D6-1, D6-4) where data were collected and annotated to distinguish sampling locations in the server room. The locations were selected based on the requirements of this research to make an initial measurement to the sound levels in dBA inside the server room.
The Quest Sound Pro Sound Level meter measures the Sound Pressure Level (SPL) for the room
associated with servers in the room computing sound pressure with algorithms at dBA, dBB, dBC, and
linear slow scales respectively. Measuring sound on the dBA scale is of special interest to the
investigator as sound for humans is primarily evaluated in a scale. This instrument is also able, however,
to measure sound in dBB and dBC scales. The dBC scale is typically used for high sound pressure levels
peaks and is virtually a linear scale as represented in Fig. 3, which includes the dBA, dBB, and dBC
curves versus relative frequencies
Experimental Study. Upon conclusion of the data compilation phase inside an actual server room, the author then performed an experimental study to evaluate and compare results. In the lab study, the recorded sound of the noise signal was generated on a personal computer with multiple speakers to simulate the ambiance of the data center in a laboratory and measure the sound levels with a sound level meter (Quest/Model 2900).
The lab study provides more control and yet a flexible environment that assisted in further defining and measuring the noise as change parameters of the noise source (i.e. the density of the noise sources and the distance between those sources) were performed. As a result, the servers inside the data center’s physical locations with different configurations could be simulated.
The lab portion of this study was performed at the St. Mary’s University Electrical Engineering Laboratory. No effort was initiated to control reverberation of sound off the floor, ceiling or furniture/equipment in the room. Pieces of software and hardware were configured to create a small lab where sound levels could be captured, modified and, lastly, analyzed for the purpose of this experiment. A MacBook Pro computer was utilized and to capture and subsequently analyze the recorded noise files. The computer was connected to a Peavey XR8600D mixer amplifier (Fig. 4) to control the audio signal of six Peavey PR15 sound speakers (Fig. 5) that were connected to the mixer through 30 ft of speaker cable. The data extracted from the results of the lab study were compared to the data. The results of this study are presented in the following section.
Site Results. In Table 2 it was included the data captured over 60-minute intervals collected in several locations inside the server room. The data in Table 2 and Fig. 6 reveal that location I-19 has the highest level of sound with 82.7 dBA, and M-18 (the location in the middle of the two server racks) is the second highest sound level of 80 dBA which indicates that the noise level was reduced because part of the signal has canceled. Front Door Hallway Corner A-28 Corner A-13 Corner W-28 Corner W-19 R-19 M-15 M-18 M-21 I-19 D6-4 Office D6-2 Office D6-1 Office 100 90 80 70 60 50 40 30 20 10 0 dBA
dBC
Experimantal Results. Fig. 7 shows a bar-chart for the average values of the sound level for the two speakers at different distances. In addition, Fig. 8 shows the real-time reading for the noise level from the two speakers. It is obvious that the level of the sound is easily can exceed the level of 90 dBA as the speakers are getting closer to each other. 91.50 91.00
At the distance of 2.2 ft and 2.3 ft, the level of the sound is lower than the sound level for the distance range of 2.4 ft to 2.8 ft. This clarify that the noise losses part of its power because at those distances the noise signal has canceled each other at specific frequencies that can be calculated by: (1) where f is the frequency (Hz) of the sound, is the wave length (ft) of the sound and sound speed is equal to 1125.3 ft/sec. Therefore, it can be concluded that a sound signal with the frequency of 490 Hz and its multipliers that is emitted from each speaker might cancel each other at a distance of 2.3 ft because of the correlation of the two sound waves. Other part of the signal can also be canceled for the same reason.
Based upon the additive law of sound, it is evident that the similarity of the noise that is generated from the servers can let to cancel part of the noise frequency as per the distance that separate between the server racks. This can help to reduce the level of the noise inside the server room by changing the distance between the server racks. 5. References