Table 1
Computational performance of our fastest A-Projection implementation described in Sects. 3.3 and 4.4.3.
| N pix | d | D | N stokes | w max | N S | N W | Δt,Δν |
|
t CF | t grid | t el | t W | Memory |
|
|
| (deg) | (km) | (s, MHz) | (h, MHz) | (%) | (%) | (%) | (%) | (Gb) | (s) | ||||||
|
|
|||||||||||||||
| 1024 | 40 | 11.4 | IQUV | 20 | 15 | 14 | 300, 0.78 | 3, 3.5 | 67.8 | 10.2 | 19.7 | 2.1 | 1 | 40.9 | |
| 4096 | 10 | 11.4 | IQUV | 20 | 15 | 14 | 300, 0.78 | 3, 3.5 | 24.5 | 4.9 | 57.5 | 12.8 | 4.5 | 117.4 | |
| 4096 | 10 | 11.4 | I | 20 | 15 | 14 | 300, 0.78 | none | 52.0 | 11.4 | 0.0 | 36.5 | 1.5 | 143.7 | |
| 8192 | 5 | 11.4 | I | 20 | 15 | 14 | 300, 0.78 | none | 32.8 | 6.1 | 0.0 | 60.9 | 6.1 | 335.7 | |
Notes. Columns list various performance parameters corresponding to
different imaging settings. The tests were performed for a 12 sub-band dataset
(1 channel per sub-band) on the CEP2 LOFAR cluster node each having 24 AMD
Opteron(tm) 6172 Processors, and 64 Gb of RAM memory.
is the
total time for a gridding or degridding step. The times
tCF, tgrid,
tel, and
tW are given as fractions of
, and
they correspond to the times needed to compute convolution functions, grid/degrid
the data, apply the element beam and apply the W-term, respectively. We are
confident we can still gain a factor of ≳ 2–4 in time over these values.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.