What I did today -instead- of Rocketry.

Wildman Rocketry
Advertise with us
The Rocketry Forum

Help Support The Rocketry Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
heada said:
Started a build of a desktop HPC cluster. 7 nodes, 28 cores, all open-source environment. I want to load openFOAM and do some CFD nozzle shape analysis.
Click to expand...

PLEASE start a dedicated thread on this -- I would love to participate and do have more than a little Beowulf in me :-D
 
Put up 12 jars of pickled okra today. Six regular, six KYD (Kill You Dead) style; extra garlic, fresh dill, fresh Serrano and Asian red peppers. Oughta cure what ails ya!
 
dhbarr said:
PLEASE start a dedicated thread on this -- I would love to participate and do have more than a little Beowulf in me :-D
Click to expand...
I didn't think there'd be any interest in that part of things. It's nothing too special, just 7x ARMv8 compute nodes (each is 4 core x 1.2Ghz Cortex A53 w/ 2G LPDDR3 each) connected via 1G ethernet and a similar for a head node. All running OpenMPI, OpenFOAM, cssh/csftp. It's very modular so I can increment compute 7 nodes at a time up to 6 modules (6 modules x 7 nodes per module x 4 cores per node = max 168 core) Each module ends up being about $400(8 port switch, 7 nodes and 7 uSD boot drives). Base OS now is Ubuntu 16.04 LTS but I'm right now in the upgrade process to get them up to 18.04.

Whole reason I'm doing this is I was reading the great Professor's book and was in the section of nozzle design. He goes over things like throat diameter, entrance and exit cone lengths, etc. I got to thinking that most of the nozzles I've seen in the hobby have the cones as true cones with straight sides. Would there be any performance difference in using convex/concave shapes in the cones? Would that change the formulas for cone lengths? Without making a TON of nozzles and testing them all out, the best way I thought of to do this was CFD on a high performance compute cluster. I'm not allowed to do personal runs on the 20k core HPC at work so I decided to build my own at home.

Haven't heard of Beowulf clusters in many a year.....good memories.
 
Pumped up the front tires on the Suburban that have been leaking since I had them installed
Rehung three pictures in the stereo room since I had it repainted.
Took the two-year-old grandson for a 2-hour power chair ride.
 
I drove Mom to the dentist in Denison again.

I'm back in Omaha now, wondering whatever happened to Fishman? Either one of them? "F*U*N Record Shop" and Fun Rockets almost seem like some sort of goofy cosmic prank.
 
NPB pharmaceutical conveyor design. Only took two and half days to do 107’ in Inventor... The guy on our team that invented it originally was a big help. Normally we have this intent code that automatically models most conveyors like E24 model then on most orders up till now you go to vault Database and like uh download parts then swap parts. Well Alt1000 and NPB you sorta scratch build without intent and it’s a major pain. And I had to design five parts today to make it do.

And the neat thing about pharmaceutical conveyor is it’s not worth the hundred million usd to make program automated to code it. So us design engineers get to hand assemble the basturd in a cad program every time an order changes specs. I quit counting after I had gee twenty three plus patterns of parts. Then there were patterns inside of patterns inside of patterns. These things get up to five lanes wide. The biggest problem is when you upload your models and basically it’s a complicated method of multiple teams of people working together to insure part numbers get named right etc. About a 2”10’ section had 78 new parts and two shafts with a 2hp Swedish motor.

Literally it had so many parts that the quad core pc at work started lagging and drawings took five minutes to load. The record is connex spirals which take fifteen minutes to load and crash our work computers often. Oh and the best part is after you design it and then the customer says they want it to fit in a personnel door... We’re like you could’ve told us that three months ago lol.

Revision control department is always pissed off about new parts. Not my problem. Lol. / Work Rant...
 
Andrew_ASC said:
NPB pharmaceutical conveyor design. Only took two and half days to do 107’ in Inventor... The guy on our team that invented it originally was a big help. Normally we have this intent code that automatically models most conveyors like E24 model then on most orders up till now you go to vault Database and like uh download parts then swap parts. Well Alt1000 and NPB you sorta scratch build without intent and it’s a major pain. And I had to design five parts today to make it do.

And the neat thing about pharmaceutical conveyor is it’s not worth the hundred million usd to make program automated to code it. So us design engineers get to hand assemble the basturd in a cad program every time an order changes specs. I quit counting after I had gee twenty three plus patterns of parts. Then there were patterns inside of patterns inside of patterns. These things get up to five lanes wide. The biggest problem is when you upload your models and basically it’s a complicated method of multiple teams of people working together to insure part numbers get named right etc. About a 2”10’ section had 78 new parts and two shafts with a 2hp Swedish motor.

Literally it had so many parts that the quad core pc at work started lagging and drawings took five minutes to load. The record is connex spirals which take fifteen minutes to load and crash our work computers often. Oh and the best part is after you design it and then the customer says they want it to fit in a personnel door... We’re like you could’ve told us that three months ago lol.

Revision control department is always pissed off about new parts. Not my problem. Lol. / Work Rant...
Click to expand...


"Listen carefully, Feyd," the Baron said. "Observe the plans within plans within plans."
 
Andrew_ASC said:
NPB pharmaceutical conveyor design. Only took two and half days to do 107’ in Inventor... The guy on our team that invented it originally was a big help. Normally we have this intent code that automatically models most conveyors like E24 model then on most orders up till now you go to vault Database and like uh download parts then swap parts. Well Alt1000 and NPB you sorta scratch build without intent and it’s a major pain. And I had to design five parts today to make it do.

And the neat thing about pharmaceutical conveyor is it’s not worth the hundred million usd to make program automated to code it. So us design engineers get to hand assemble the basturd in a cad program every time an order changes specs. I quit counting after I had gee twenty three plus patterns of parts. Then there were patterns inside of patterns inside of patterns. These things get up to five lanes wide. The biggest problem is when you upload your models and basically it’s a complicated method of multiple teams of people working together to insure part numbers get named right etc. About a 2”10’ section had 78 new parts and two shafts with a 2hp Swedish motor.

Literally it had so many parts that the quad core pc at work started lagging and drawings took five minutes to load. The record is connex spirals which take fifteen minutes to load and crash our work computers often. Oh and the best part is after you design it and then the customer says they want it to fit in a personnel door... We’re like you could’ve told us that three months ago lol.

Revision control department is always pissed off about new parts. Not my problem. Lol. / Work Rant...
Click to expand...
Sounds like you could use some more RAM and a better drive. Been dealing with some pretty large assemblies in SW lately and just bumped up to 64GB ram and installed an NVMe drive. I've got an i7 7700k and WX8100, so it didn't seem to be CPU or graphics related. It's night and day. I was having a ton of issues with crashing previously and HSMworks doesn't autosave... I got pretty sick of forgetting to save and losing a few hours of programming every time SW would lock up. :confused:
 
Oddly I need more storage space only worked a few months and used 2.3 TB. The level of detail was the issue with connex spirals not ram. The program couldn’t handle a complex structure 55 ft tall, 20 ft wide, and with eight hundred feet of support tubing and joints around it. We had hit a polygon limit with Autodesk inventor. And it was made worse when the customer requested it fit through a house door because then we had to design crap in very tiny pieces to fit through said door. They have to fly people from Utah and Sweden to Arkansas, train them how to assemble it, and help set it up in the field. One of our solutions was to vary LOD for certain stuff. It’s bad when you load a drawing and the base view is just a blob of black lines everywhere. Oh and the Swedish company of careyline connex conveyors has never ever designed their product to fit through a small door. Needless to say their horrified and I’m certain this hasn’t been done before. We just feel bad for the assembly guys. They’ll need to hang off a literal ceiling lol.
 
heada said:
Whole reason I'm doing this is I was reading the great Professor's book and was in the section of nozzle design. He goes over things like throat diameter, entrance and exit cone lengths, etc. I got to thinking that most of the nozzles I've seen in the hobby have the cones as true cones with straight sides. Would there be any performance difference in using convex/concave shapes in the cones?
Click to expand...
I've often wondered the same, and after some small time searching have found no answers. What about somethng like this?
upload_2019-9-26_12-54-38.png
Y=±[1.5+cos(x²)], right end slightly truncated. There are no sharp spots.
 
jqavins said:
I've often wondered the same, and after some small time searching have found no answers. What about somethng like this?
View attachment 394289
Y=±[1.5+cos(x²)], right end slightly truncated. There are no sharp spots.
Click to expand...
My understanding is it's a couple percentage off theoretical peak, IFF you have a fully combusted gaseous exhaust. Otherwise the various solids glom up the curves.
 
Finally pulled the rear wheel off my mountain bike. It is tubeless, but wasn't holding air. Turned out to be a crack in the aluminium rim, inside :(. I'll see what the bike mech says, but I might also give it a go with some TIG welding. Alternatively I am up for a new rim ($$$).
 
Welding it will throw the wheel out of balance unless you're really good.

Anyway, changed one tail light bulb and both headlight bulbs on the pickup yesterday. Only one headlight was out, but I did the good one first; D'OH! And those headlights (2003 Ram 1500) and a real PITA.
 
jqavins said:
I've often wondered the same, and after some small time searching have found no answers. What about somethng like this?
View attachment 394289
Y=±[1.5+cos(x²)], right end slightly truncated. There are no sharp spots.
Click to expand...
That's exactly where I was thinking. If a curved shape was ideal and if so, what curve and lastly how would it react in practice vs theoretical. The computer part will only handle the first 2 questions. Once I have an ideal shape, I'll see how hard it is to make in both graphite and re-enforced ceramic. I've turned a graphite nozzle once......once. Since graphite is so nasty to turn, I also want to see if you can use ceramic with refractory fibers embedded and then glazed as a substitute for graphite. It'd be slightly less durable than graphite but much better than phenolic but can be shaped/molded like phenolic. My gut says it'd reduce heat-soak in long-burn motors too but I have no basis for that other than gut-feel since ceramic is a fairly decent insulator.

I had to put the cluster on the back burner for a few weeks. Got busy and my management at work got wind of what I was doing and is wanting me to bring it in and make it an official project. They even offered to let me expense the hardware I've gotten so far. Not going to take them up on expensing it since I want to own it once its done and working but I think I will bring it in and I can work on it during slack time at work. I've since purchased 2x 5v 60A power supplies, a few DC boost converters (5v step up to 9v for network switch, 5v step up to 12v for fans if I need them), 5v DC cooling fans and I'll be building a case over the next week or three. I have the redundant power and connections to support up to 28 compute nodes, 112 cores and all in 1U of rack space.
 
Didn't you say earlier that you're building your own cluster because you can't use the super one at work? If they want you to make it an official project then use their equipment.

Anyway, as to making them, if you wan to try ceramic then how about this: turn a negative in wood, cast the ceramic around it, then burn the wood out. Or, turn it in the material of you choice, but in two pieces with the break right and the throat.
 
jqavins said:
Didn't you say earlier that you're building your own cluster because you can't use the super one at work? If they want you to make it an official project then use their equipment.

Anyway, as to making them, if you wan to try ceramic then how about this: turn a negative in wood, cast the ceramic around it, then burn the wood out. Or, turn it in the material of you choice, but in two pieces with the break right and the throat.
Click to expand...
The idea was to make a 2 part mold and cast ceramic around it. I have a friend with a kiln that could fire a dozen or so 75mm nozzles at once. It'd take a day or two to fire and cool but its all automated. Stone-age tech in space-age hardware!

I can't use the 20,000 core cluster at work because it's used 24/7/365 to do pharma/genetic sequencing research. I'm sure I could make a case to use it if I could offset the cost.... more than $50k/minute. The idea behind making this an official project is that if I can make it work for less than $500, I can put one on each scientist's desk and let them do a proof of concept run and if it passes, then they do the full run on our big HPC in the data center. The 7 node, 28 core setup uses less than 100W of power so it's easily a desktop solution rather than the 37kW plus chilled water cooling per rack in the data center. And I'm only the storage guy, not the compute guy so I don't even have access. I manage the 50PB of storage at this site, 300PB globally.
 

Similar threads

Wayco
Wayco and Hardline's Northern exposure
Replies
13
Views
450
Wayco
Wayco
DAllen
Show me your multi-tools!
2
Replies
33
Views
2K
woferry
W
StreuB1
30th Anniversary
Replies
1
Views
347
Hobie1dog
Hobie1dog
B
anybody use pinstripe tape?
Replies
19
Views
674
lr64
L
Y
Introduction
Replies
18
Views
523
Yoda
Y

Latest posts

Back
Top