Grid Computing Center

Gilthanis

Gilthanis

[H]ard|DCer of the Year - 2014
Joined
Jan 29, 2006
Messages
8,761
So, I set up a 20GB Ramdisk to put a ton of multi clients into. Loaded up the first 700 in 100 increments and it kept the full 8 total cores at full load... And they were erroring out. Will try loading up in 100 increments again but checking them with each step to see how it goes. I did the ramdrive to eliminate the IO potential just in case. Been trying to see what limits I have with this application. I am wondering if it is a CPU limitation even though it is nci...
 
The last time with goofy, I can only max 250 clients per rig. I'm currently running 220 clients without error for a week. I'm itching to increase this but let see how you are doing first, lol.
 
Well...it didn't like 700 clients. It kept all 8 cores pegged the whole time. I'm gonna start with 100 and then move to 200 and so on as I see how they succeed/fail.
 
I have to hold off a bit... I was playing around with the Ramdrive which I hadn't messed with much. If I set it to use 20GB ram and then tell it to save everything to an image file on the HDD, it doubles the amount of RAM it uses. But, if I don't have it save to an image, a restart wipes the files and starts over. I have it currently running burstcoin on the ramdrive for experimentation. However, as soon as I have enough time to set it up again, I will start with the multiple clients again. I just wanted to see how far a single rig could really push things. And I still haven't figured out why it craps out other than maybe being a CPU limitation due to possibly client overhead in combination with the apps. 700 certainly pegged 8 full cores. That would be 2800 work units plus 700 boinc.exe instances running....
 
Do you think virtual box approach might work? Haven't tested it yet. Trying to setup another rig,, heh, heh.
 
Depends on the rig, but I was able to run around 200-250 depending on the setup. However, XP VM's really didn't like more than 100 each. I recommend VM's because if they start stealing cores or begin erroring out, they don't steal or mess with other work units. Also, if you have a bunch start erroring out, it is easier to shut the VM down than the whole system. This Ramdisk situation was just to see if it helped any, but my first experience is CPU time for all of those clients. You have to stagger them too as they will want to do a quick benchmark...

I just started 200 clients back up on one of two ram drives. The other ramdrive is running burstcoin to see how it does with pure memory vs. HDD speeds.
 
Last night, I just increased the number of clients from 220 to 250. So far no error on any tasks after half a day of running. If it is stable, I'll add one more client tonite and see if there are errors in some of the tasks.
 
Forgot to document this. 250 clients (1000 tasks) running with no error....

upload_2017-6-27_12-56-3.png
 
200 ran fine on the Ramdisk, so just bumped up to 300 on that rig.
 
And...sure enough... errors... it may have a limitation of 250..... :(

My VM's are down as I'm in transition of data right now. Can you try running more than 250 on a box by using more than one VM to see if it is a box limitation or an OS limitation by chance? I know on XP I was limited to like 100 and sometimes up to 150 but not consistently. I'm wondering if the Win 7/8/10 running more is because I've ran them as 64bit. It may be limitation somehow. Just a theory...
 
Last edited:
I fired up two more clients (252 or 1008 tasks) and got error within 20 mins or so. Yep 250 is the hard limit. No change from last time before it moved to GCC.

Let see if I should play with vb.

upload_2017-6-27_20-40-43.png
 
Yes...please start up a VM running 250 on the primary host which actually would probably be 251 counting the one that starts with boinc manager. Then see if you can also run some more within the vm as well without errors. IIRC, I was able to but I had so many different things going and using the VM's at once, when I did get errors I was never consistent with finding what caused it. Just not enough time right now to do a lot of testing and I'm not a fan enough of the science to put too much into it. However, it is easy points for those wanting them.
 
Success. After running overnite, I'm able to fire up another 250 clients on the same rig with VM approach. I've 500 clients running now on a single machine (2695v2). I'll monitor this for a few more days. For VM, I allocated two cores and they do use a bit of CPUs during tasks end/ new tasks start, I believe. I think I can setup another two more VMs on this rig but then I would probably reach a limit on my memory and SSD space.


upload_2017-6-30_7-31-52.png
 
That is what I was doing on a smaller scale before. Also, are you VM's using 64bit OS? I'm just curious as I haven't tried a 32bit version of 7 or 10. I'm wondering if it is 32bit limitation I was having with XP or if it was just XP in general.
 
Sad part is that I don't care enough to actually load a 32bit Win 7/10 VM up to test it. I am curious though if it is a limitation of the way we are doing it or if there are other restrictions. Like, if someone used Docker in Linux.....
 
Since BURST is currently a waste of time for me, I'm gonna start playing with more GCC VM's and see how far this POS server can go. Gonna add a couple more VM's to the same HDD running 250 clients each. See if that is happy. If so, may create another VM on another HDD...etc... The box has 48GB Ram in it total to play with and can have up to 26 HDD's to spread the IO's across. However, I'm not sure yet what plan I want to go with as it only has a total of 8 cores to feed the VM's and not sure it could handle more than a half dozen. Oh well... you never know until you try.
 
You must log in or register to reply here.
Back
Top