Estimation of resource for OSSEs
gsf T382 |
Moorthi Dew/Mist |
104 task, 13 node Th=2 |
8.5min /day 60min/ 7 day |
gfs T382 3 hourly dump |
Jim Jung Vapor |
16 task 1 node Th=4 |
20min for 6hr |
gsf T382 |
Jim Jung Vapor |
16 task 1 node Th=4 |
3.5 hour/7day |
gfs T382 |
reanl Dew |
224 task, 14 nodes, 1 thread |
20 min/9hr 4hr20min/5day (6 hr/7 day) |
gfs T126 |
reanl Dew |
128 task 4 node Th=1 |
25 min /5day 35 min/7day |
gfs T62 |
reanl Dew |
16 task 1 node Th=1 |
1 min/9 hour 9 min/5day 13min/7day |
GSI T382 |
Operational analysis Russ Treadon Dew |
160 task 10 nodes (16 tasks per node) th=1 average run
time: 1146.446 seconds |
20 min/cycle 80min/day |
GSI T382 |
Jim Jung haze |
64 task 4 node TH=1 |
one hour/cycle 4 hr/day |
GSI T382 |
Jim Jung Vapor |
32 task one node real cpu only Airs and IASI |
one hour/cycle 4 hr/day |
GSI T382 |
reanl Dew |
224 task 14 nodes Th=1 |
25 min/cycle 2hr/day |
GSI T62 |
reanl Dew |
16 task 1 node TH=1 |
10 min/cycle 40min/day |
Simulation of Radiance |
Tong Zhu Haze |
one cpu GOES-E/W, AMSU |
0.5 hour/day |
Simulation of Radiance |
Ron |
one cpu HIRS, AMSU, AIRS no GOES |
3 hour/day |
|
|
|
|
[Summary of estimate on vapor]
In general vapor is 30-50% faster than haze.
Rough estimate to process one day of assimilation with two five day forecast
T62
one node
GSI 30min/day (4 cycle)
two 5day fcst 15 min
45 min/day one node
T126
2 node
GSI 40min/day
two 5day fcst 20min (rough guess from computation with 4 node)
1 hr/day 2 node
T382
one node
GSI 4 hr/day (4cycle)
one node
two 5day fcst 5 hour
9 hr/day with one node
T62
Number of experiments. 100 exp for one year =36500 day
1(node)x0.75(hr/day)x36500(day)=27400 hr*node=1142 day*node
analysis only 200 exp for one year =73000 day
1(node)x0.5(hr/day)x73000 (day)=36500hr*node=1520 day*node
T126
Number of experiments. 100 exp for 6 week =8400 day
2(node)x1(hr/day)x8400(day)= 32000 hr*node=1333 day*node
20 experiments for one year=7300 day
2(node)x1(hr/day)x16000(day)= 32000 hr*node=1333 day*node
T382
100 exp for 6weeks=4200days
1 nodes 9 (hr/day)x4200 day = 37800 hr*node=1575day*node
200 exp 6wk with T126
20 exp one year with T126
20 exp 6wk with t382
6 hourly cycle
fcst twice a day
2 node for 860 days
do not include resource for debug
Simulation
existing radiance data
365x3 hr (existing + future=1095 hr x 1 task
34 hrx 64 task
If coefficient has to be calculated, it will take more time.
Time for simulation is negligible.
Each day produce 1GBfile
Run history of 15000day 15 TB
GOES-R
20km 10 times thining -> 60 km resolution : 15MB/hr 360MB/day=150GB/year
20km 3.6GB/day=1.5 TB/year
4km 3.6x25=90GB/day=33TB/year
GOES-R CrIS, ATMS
We have tom mix 4km 20km and 60km resolution
50TB hpss space to save the simulated data for new satellite
50TB hpss space to save results
10TB disk space