Velvetg: 'Killed' During ___/___ Nodes Visited

Question

[Jared]

I've been trying to run velvetg but everytime it returns 'Killed' after nearly getting through all of the 'nodes visted' part, for example:

[11019.929049] 11860000 / 11927480 nodes visted
Killed

I have sufficient hd space available and I don't know where to proceed because of vagueness (and frankly rudeness) of 'Killed'. Thanks in advanced.

edits: data is short paired illumina reads

k value=21

insert length=150

minimum contig length=100

read trackage is on (data is a transcriptome that I want to run through Oases afterwards)

as far as memory goes, as I'm running velvetg again top is showing about 10% memory use

Edit 2: Thanks for all the advice! I think I just don't have enough memory to muscle through this with the current settings, I upped the kmer value and I've had much more success.

[David Quigley]

Programmers routinely (if rather insensitively) refer to terminating the action of a running process as "killing the process". "Killed" is not a helpful error message to the end user, but I wouldn't take it personally. Don't even get me started on "master and slave processes", though...

[Marina Manrique]

what is the length of your reads and the k value you're using?

[Neilfws]

The system will kill processes that consume too much memory. Have you monitored memory usage during the run (e.g. using top) ?

[Marina Manrique]

maybe you're using a k too small. It's recommended to set the k at least as the minimum of the read length

Answer 1

[Alex Richter]

In general, if you are running a program in Linux, and it exits with a 'Killed' message (and assuming that you haven't kill -9'd it yourself), it means that either you hit the ulimit (as per Istvan's comment above), or that the kernel was forced to kill it due to running out of memory (both physical and virtual) See http://stackoverflow.com/questions/726690/who-killed-my-process-and-why for an overview.

Velvet is nothing if not an incredible memory hog. See http://seqanswers.com/forums/showthread.php?t=2101 for a quick algorithm to estimate how much memory an assembly will require.

As a point of reference, I am working with high diversity metagenomic populations, and a single lane of Illumina Hi-Seq requires upwards of 800GB of RAM. If I run it on a machine that only (!) has 256GB of physical RAM, it takes over two weeks to complete, where if I run it on our high memory server, with 1 TB of RAM, it finishes overnight. The time difference is almost entirely due to the server swapping memory out to disk.

Answer 2

[Istvan Albert]

This sounds like exceeding limits set on the account via ulimit. The bash shell is killing the process. Check your limits with::

ulimit -a

Also read up on ulimit to see how to change it.