32 bit Synthetic GOESR and NPOESS Imagery 1
32 -bit Synthetic GOES-R and NPOESS Imagery 1) 1997: AIX operating system, 196 Mbytes of ram, few hundred Mbytes of disk space, RAMS-3 b. 2) 1998: Linux operating system, 512 Mbytes of ram, 500 Mbytes of disk space, RAMS-429. Systems administration was done by user. Wow! Just think what I can do now. 3) Upgrades! 1 Gbyte of ram 2 Gbytes of ram 3. 3 Gbytes of ram. Standard Linux Kernel Bigman Kernel Experimental Kernel. Disk space increased to 100 Gbytes. RAMS-43. Systems administration done by systems administrator—not the user. Oh boy! Just think what I can do now. 4) 32 -bit cluster, 20 dual processor nodes, 2 Gbytes of ram, few hundred Gbytes of disk space. Test RAMS-43. Gee-wee what a fast machine! Just think what I can do now. GOES-12 10. 7 µm Louie Grasso, Manajit Sengupta, Jack Dostalek, and Mark De. Maria 6) Test RAMS 43. (a) Reproduce a 32 -bit run, (b) Exceed 4 GB of ram. 7) Repeat (6), this time do runs in 1) Purchase a ready-built machine or parallel. 1) Run RAMS 43 and use output to make build one? We built one ($$$). 8) More ram added: synthetic GOESR-ABI and NPOESS 2) How much ram should be put in 8 GB 16 GB 32 GB is VIIRS images for three mesoscale weather the machine? Started with 8 GB. hardware limit. events. 3) How much disk space? Started 9) More disk space added: 2) 200 Gbytes of disk space should be with 500 GB. 500 GB 1 TB 2. 5 TB raid to plenty—I was in for a surprise! 4) How many nodes and what 10 TB raid. 3) 32 -bit 4 Gbyte ram too restrictive. processor speed? 10 dual processor 10) Each time more ram was added, GOESR-ABI has 2 km footprint, but I had nodes. I re-ran the same job to take to use 4 km horizontal grid spacings. 5) Get RAMS 43 to run on the 64 -bit advantage of the extra memory by NPOESS VIIRS has a horizontal footprint cluster. This took six week. The increasing the horizontal domain. near 400 m, but I had to use 1 km issue was this: RAMS 43(64 -bit) but 11) The first time I submitted a 25 horizontal grid spacings. mpich(32 -bit). That is, the 64 -bit GB job, a lot of time went by before 4) Horizontal extent of domain too small to rams variables had to be changed to the model wrote out the initial data. I produce a realistic looking synthetic 32 -bit variables before MPI thought something was wrong, so I satellite image. subroutine calls then changed back did what any good scientist would 5) Must move to 64 -bit cluster: A whole to 64 -bit after the suboutine. do—kill the job! new world! 12) Fifteen minutes passed before GOES-R 10. 35 µm NPOESS VIIRS 11. 02 µm RAMS 43 wrote out 15 GB of initial data. I had no idea the files would be this large. NOAA/NESDIS Co. RP, 15 -16 August 2006, Hilton Fort Collins, Colorado GOESR and NPOESS 64 -bit environment 1) Where should the machine go? 2) Fuses to room are blowing! 3) Power outages! 4) Air conditioning? 5) Room too hot! The machine is overheating! 6) Machine damaged! 7) Results are expected, and quarterly reports are still due. 8) Machine rebuilt; new systems administrator. Hurricane Lili Lake effect snow GOES-R 6. 185 µm GOES-R 9. 61 µm Severe weather GOES-R 10. 35 µm GOES-R 13. 3 µm
- Slides: 1