|
From: | Eric Melski |
Subject: | Re: single-threading targets although make was invoked with -jX ? |
Date: | Wed, 4 May 2011 10:26:12 -0700 |
User-agent: | Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.2.14) Gecko/20110223 Thunderbird/3.1.8 |
On 05/04/2011 10:20 AM, Eric Melski wrote:
On 05/04/2011 09:47 AM, David Boyce wrote:On Wed, May 4, 2011 at 11:20 AM, Per Jessen<address@hidden> wrote:yeah, SYNCSH_SERIALIZE is equal to what I do today using this construct: (flock -s 200; some-command $^ $@) 200>/var/lock/some-lockfile It satisfies the single-thread requirement, but in massively parallel runs, several of these often end up waiting for each other.Could you rephrase this? I can't see a meaningful distinction between "waiting for each other" and "submit one at a time". Either way they're serialized, no?They are serialized either way, but the processes could be consuming other resources (memory, disk, database connections, etc). In the "submit one at a time" model, you avoid tying up those resources needlessly.
Oops, should have had my coffee before responding. :) I see now the synchronization occurs outside the invocation of the process, so the process won't be able to acquire any other resources, because it won't actually have started.
I suppose if you're looking at a really, really massively parallel system you might have concerns with swamping the process table with effectively idle processes though.
br, Eric Melski Architect Electric Cloud, Inc. http://blog.melski.net/
[Prev in Thread] | Current Thread | [Next in Thread] |