Professional Documents
Culture Documents
Keywords: aisle containment, hot air recirculation, air flow distribution, modified body
force model, porous jump model, server inlet temperature
Fig. 2 Comparison of PIV and CFD results for the fully provisioned case
and 2(a)). The aisle top is also open which will allow for air Fig. 1(c)). The porous jump model is also not able to capture the
by-pass or entrainment. For further details of the computational air entrainment from aisle center, near the tile surface (bottom left
set up, geometry, and operating conditions, refer to Ref. [7]. corner) and the delayed entry of air in the lower portion of the
From the experimental data shown in Fig. 1(b), we note that rack inlet (bottom right corner). The porous jump model also
even though the air flow rate through the tile is lower than the suggests entrainment from the aisle top which is contrary to the
rack air requirement (under provisioned, case 1), a portion of the experimental observation.
supplied air is observed to escape from the aisle top. We can also Flow field with the geometrical resolution model that included
observe the delayed entry of the tile air at the lower portion of the the geometrical details of the tile and with the modified body force
rack inlet (see bottom right corner of Fig. 1(b)). This was attrib- model is shown in Figs. 1(d) and 1(e), respectively. From the fig-
uted to the higher momentum of air emerging from the tile [7]. ure we observe that the flow field features for both of these models
Air entrainment from the aisle center can also be noticed for this are closer to the experimental results. A portion of air flow is
case (see bottom left corner of Fig. 1(b)). The porous jump model observed to escape from the aisle top as observed from the experi-
is not able to capture these features and the air flow from the ments. These models also capture the delayed entry of air to the
tile is predicted to reach only up to a certain height of rack (see lower portion of the rack inlet (bottom right corner). The air
Fig. 3 Comparison of geometrical resolution, modified body force and porous jump model for middle racks in a contained aisle
for the fully provisioned case in contained aisle condition
Fig. 6 Comparison of open and contained aisle for middle racks (4,11) with modi-
fied body force and porous jump models
Fig. 7 Comparison of open and contained aisle for end tiles (1,14) with modified
body force and porous jump models
present case temperature rise of 10 C is specified corresponding The velocity contours along the center plane crossing the end
to a rack heat load of 15 kW with rack air flow rate of 1.224 m3/s tiles 1 and 14 is shown in Fig. 7. For the MBF model with open
(2594 CFM). Region inside the racks, CRAC units and PDUs aisle, the flow field is very different as compared to that corre-
is not considered and no mesh is present in it. The modified sponding to the middle tiles (compare Figs. 6(a) and 7(a)). This
body force source term is specified in the region of may be due to air entering the cold aisle near the bottom portion of
56 cm 56 cm 15 cm (22 in. 22 in. 6 in.) above the tile sur- the aisle. This can be seen as high velocity region near the bottom
face for all the tiles, see Fig. 5(d). of the aisle, see Fig. 7(a). This can be further understood from the
Figure 6 shows the velocity contour plots for fully provisioned temperature contours shown in Fig. 10 later. For the contained
case along the center plane crossing the middle tiles 4 and 11. aisle, the velocity field at the middle and the end of the aisle is simi-
From the figure we observe that the flow field is almost similar for lar due to absence of air entrainment; see Fig. 7(b). With PJ model
both open and contained aisles. The modified body force model the velocity field is similar for both open and contained aisle, sug-
predicts a low velocity (V-shaped) region near the aisle center gesting much lower entrainment as compared to MBF model.
which is absent for the porous jump model (PJ) results. The flow Figure 8 shows the temperature contours at the center plane
field is similar to that calculated for the case simulating middle along the length of the cold aisle for the under provisioned case.
racks in a contained aisle as shown in Fig. 3. The balance rack air requirement is met by hot air recirculation
Fig. 9 Comparison of (a) modified body force and (b) porous jump model along
the center plane of partially contained cold aisle (only the top curtain deployed)
and under provisioned case
from the room. From the figure we can observe that MBF model may be because the top curtain can curtail the hot air rising and
suggests significantly higher hot regions as compared to the PJ escaping from the aisle top and the entrainment effects are
model, possibly due to higher air entrainment. This result suggests suppressed.
that the PJ model may underestimate the server inlet temperatures The temperature contours for the fully provisioned case is
and can lead to flawed air flow pattern design. shown in Fig. 10. From the figure we note that, even though the
With partial containment, where only the top curtain is cold air supply and demand are equal, hot regions are present sug-
deployed while the aisle doors are opened to allow free access to gesting hot air is entrainment into the cold aisle (see Fig. 10). The
the cold aisle (refer to Fig. 4), the temperature field suggests MBF model suggests much higher air entrainment as compared to
improvement in cold air supply scheme. There is minor difference the PJ model similar to that observed for the under provisioned
between the MBF and PJ models for this case (see Fig. 9). This case (also, see Fig. 8).
Note that in this case no net air entrainment can be present inlet temperatures corresponding to the six blade center for each
because the supplied air is matched with the rack air requirement. rack (see Fig. 5(b) for more details). Each bar in the figure
For this case also the MBF model suggests higher gross entrain- shows the mass weighted average temperature of a modeled
ment as compared to the PJ model. This investigation shows that blade center inlet. The bars from left to right represent servers
for the fully provisioned case, partial containment can be an effec- from bottom to top.
tive means to supply cold air to servers and curtail hot regions in From Fig. 12 we observe that there is a large variation in server
the cold aisle. inlet temperature (25 C) for the under provisioned case. The
The entrainment rates through the open doors for various cases MBF model suggests much higher server inlet temperatures as
investigated here are given in Table 3. For open aisle, we observe compared to the PJ model; possibly due to higher air entrainment
that the hot air entrainment rates are significantly higher with the rates (see Table 2). The results also suggest that the use of PJ
MBF model as compared to the PJ model. For the under provi- model may not provide appropriate prediction while designing air
sioned case, with the MBF model, the entrainment through the flow schemes. Use of partial containment suggests reduced server
doors is 36%, whereas, with the PJ model it is 24%. With par- inlet temperatures with MBF model. This shows that the use of
tial containment the balance air has to enter through the open partial containment can be an effective for cold air management
doors, hence the entrainment rates for the PJ and MBF models are to reduce high server inlet temperatures. This difference is not no-
similar in this case (40%). ticeable with the use of PJ model.
For the fully provisioned case, with both open and partially For the fully provisioned case, much lower server inlet temper-
contained aisles significant gross entrainment rates are present. atures are observed (see Fig. 13). With open aisle, due to hot air
However, note that with partially contained aisles the air mostly entrainment, the server inlet temperatures are higher for the end
enters through the bottom portion of the open door and escapes racks (rack# 1, 14, 7, and 8). Refer to Fig. 5(a) for the layout of
through the top portion. Hence the influence of the hot air on sever the data center. In this case also the MBF model predicts higher
inlet temperatures is expected to be minimal. server inlet temperatures as compared to the PJ model. With use
ASHRAE has recommended server air inlet temperature range of partial containment the server inlet temperatures are signifi-
from 18 to 27 C [14] for effective operation of the servers. To cantly reduced and the variation is within 5 C suggesting effec-
meet this recommendation, lower variation in the server inlet tive air delivery scheme for this case.
temperatures as well as appropriate supplied air temperature is
required. However, with higher hot air recirculation the tempera-
ture range may be higher, posing challenge to meet the ASH- 6 Conclusions
RAE recommendation. Figures 12 and 13 shows the computed Effective air delivery schemes require appropriate choice of
server inlet temperatures for the cases with under provisioned computational models to accurately predict the flow and thermal
and fully provisioned cases, respectively. Note that each rack is fields in a data center. For a single rack, it was observed that the
modeled to have six blade centers and hence there are six server modified body force model closely predicted the major flow