torptube
commented
4 years ago I would recommend not using swap at all, and turn the NVMe drives into a RAID-0 array for scratch space holding the database and files to be processed. If your databases aren't on a high speed I/O device, the blastn steps could be significantly impacted.
You can use what the installer does to make a RAID-0 array on any system with more than one local SSDs using this:
sudo mdadm --create /dev/md0 --level=0 --raid-devices=$(sudo nvme list | grep "Amazon EC2 NVMe Instance Storage" | wc -l) $(sudo nvme list | grep "Amazon EC2 NVMe Instance Storage" | cut -f1 -d " " | tr "\n" " ")
sudo mkfs.ext4 /dev/md0
sudo mount /dev/md0 /scratch
Those commands should make a RAID-0 array with whatever local SSDs you have and mount it to /scratch.
Regarding your error: it's not a memory error, you have to raise the open file limit, and I would raise the amount of threads too, just in case.
You can change the limits using ulimit:
ulimit -a to find out your current limits
ulimit -n 65536 to set max open files to 65535
ulimit -u 4096 to set max threads to 4096
Every system is different as to the hard limits, I don't remember the hard limits on Amazon Linux 2, but I think they are higher than the previous suggestions.
There is no ability to resume from a previous run. Sorry. However, depending on what you are looking for, you could potentially skip the time intensive steps of megahit and iterative improvement. Try running with and without both/either --noAssembly --noIterImp and see if the results are similar enough such that you can just skip one of both of those steps.
jingzhejiang
Hi,
I run virmap on c5d.24xlarge (96 vCPU, 192 GiB, 4* 800Gb) instance with an 838Gb SWAP setting
It works for many libraries, but didn't for another similar size libraries (maybe with too many viral contigs ? I'm not sure.) For the error libraries, they threw out many empty files after
12S-WGA.filtered.fastep.Here is the record of error log file
I also checked the end of file 12S-WGA.filter.err.bz2, and found error messages at the last line:
Is this a memory error, even after an 838Gb SWAP is set? Can I overcome the problem of memory error by adjusting SWAP setting, or I can only change to some expensive, big memory instance?
ANOTHER IMPORTANT QUESTION is: Can virmap continue to execute from the interrupted steps, thus saving some time and money?
Thank you for your reply!
12S-WGA.err.gz 12S-WGA.filter.err.gz