LISTSERV mailing list manager LISTSERV 15.5

Help for LINUX-L Archives


LINUX-L Archives

LINUX-L Archives


View:

Next Message | Previous Message
Next in Topic | Previous in Topic
Next by Same Author | Previous by Same Author
Chronologically | Most Recent First
Proportional Font | Monospaced Font

Options:

Join or Leave LINUX-L
Reply | Post New Message
Search Archives


Subject: Re: Anyone interested in a server build contract?
From: John Gelnaw <[log in to unmask]>
Reply-To:Platform Independent Linux List! <[log in to unmask]>
Date:Thu, 24 May 2018 13:20:40 -0400
Content-Type:multipart/signed
Parts/Attachments:
Parts/Attachments

text/plain (37 lines) , signature.asc (487 bytes)


On Thursday, May 24, 2018 12:59:27 PM EDT Kozakoff,Stephen J wrote:
> I would be curious if building in memory might improve performance. I know
> tmpfs is subject to being swapped out to disk, but, it's got to be faster
> than just straight disk if there is enough memory in the box.
> 
> Something like...
> 
> mount -o size=1G -t tmpfs "build" /mnt/build

Unlikely, based on the phoronix tests. Also, since it's the Android OS, it's 
kinda huge.  Finally-- in the pet peeve category:  Stop using /mnt for 
anything that isn't transitory.  :)

I suspect the biggest improvement would be gained by using:

# make -j`nproc`

instead of just "make".  Anecdotally, on a box with 12g of RAM, it apparently 
reduced the build time from 6 hours to 40 minutes (according to j random 
internet poster.  Apply salt as needed).

Personally, I would recommend a 16 (or more) core server with at least 32gb of 
ram, tiered storage (RAID-5 for long term for storage, one or more SSD's for 
build/scratch), and a copy of gitlab community edition with continuous 
integration-- in short, an actual build server, and stop abusing the laptops.

Do the builds inside a docker container (there are several that already exist) 
to maintain the ubuntu 14.04 environment, and use as much memory and threads 
as possible.

In short-- Boost everything except the storage bandwidth, because it probably 
isn't the limiting factor.  :)

-- 
John Gelnaw <[log in to unmask]>
University of Florida Health IT

Back to: Top of Message | Previous Page | Main LINUX-L Page

Permalink



LISTS.UFL.EDU

CataList Email List Search Powered by the LISTSERV Email List Manager