Friday, August 23, 2013

24 core cluster slaughtered (for the time being)

The HPC cluster I built some time ago consist(s/ed) of 4 ASUS MA58L LE motherboards, having 4GB of RAM and an AMD FX-6100 processor. With one head node and three worker node, this summed up to a total of 24 cores with 16GB of RAM, all linked up through Gigabit. Depending on the needs, i.e. OS, of the time, the workers were equipped with a 1TB HDD, the head node's storage was/is 1TB too.

As things develop, I wanted to secure my precious data, held on my tiny NAS, a bit better. I decided that a second RAIDZ-1 NAS would be in order.

Second RAIDZ-1? Yes indeed. My first RAIDZ-1 file server is based on an Acer easyStore, which came with windows home server, and now runs nas4free. The reason it runs this particular system is the built-in SSD, see earlier posts why. The easyStore is a really weak Intel Atom based server... nothing wrong with that, however, Sun's ZFS may be a bit of a challenge for this particular architecture.

Here is idea, use the 4 1TB disks in the former head node and add another 4GB to it. A 6 core processor, somewhat more powerful than an Atom, running disks in SATA3 and having 4 times more RAM, should make a certain difference. This will hopefully not only enable the "dedup"-option, but also render the "scrap-NAS" (see earlier post) superfluous.
Potentially, if things work out well, the disks of scrap-NAS will be inserted into the new 6 core device.

The next step would be, to build another head node for the HPC cluster, I figure.



No comments:

Post a Comment