Deconstructing my build hive... As you may tell form earlier posts this month, I'm on a mission to sort out some of my equipment and either recycle it or whatever, and on this mission one of the first stops has been the older dual socket servers, which live in the garage on an ethernet spur at the moment.
And, I've taken them down, now I didn't really appreciate how much they lent to my build times, as my main app server (on which I coordinate the building of gcc projects with distcc) started to report 4x longer builds (on average) and I couldn't find a reason for it.
I looked at the app server, I looked at the jobs being thrown about and really there was no reason for the massive slow down...
Until I remembered something about my VDI set up, the app server used two sources for it's virtual disk images, one local and one was passed over to the containers on the garage servers, these images could be mounted over NFS but more regularly I had the garage servers sync them and mount locally from their local disks.
In migrating the build containers back locally to the app server, I'd moved the containers from their local loop mounted VDI's to ones hosted on the app server... From being mounted from 15,000 RPM SAS disks at about 12gb/s to a SATA 7,200 RPM disk at 6gb/s.
I'd essentially halved the disk speed of my builds, and totally forgot about setting this all up, now I did this years ago, it must be about two years since I set it up and it's been nearly a year since I moved house and started everything back up on the ethernet spur, which necessitated using the copy/sync and local mount option to get over the slow 10mbs ethernet in the garage.
My plan therefore is to get the new power supply, experiment with the supermicroo motherboard and then see if I can set up a new ZFS based file server on that SAS controller... Watch this space.
No comments:
Post a Comment