the volume storage solution
Application:
So this is for a development video encoding platform.
Reads can be really slow (specifically, the product of the peak number of simultaneous clients * the max bitrate of blu-ray 1080p MPEG-2’s, which is 40Mbps)*[http://www.blu-ray.com/faq/] (And, not sure if seeking bumps this number up significantly). Realistically, we’re streaming h264 content, so reads aren’t going to task the system on reads… and data will be written so we can leverage extents to avoid fragmentation…
Writes - the data is going to come from post-processing efforts from tier1 storage so it should be able to keep up with the generated content, but again, this is probably well within range of the max write speed of even a single disk at any given time
Initial size matters less; expandability and data integrity is more important overall.
Firstplan:
sil3132 + sil 4726 5 port multiplier, raid6/10/50? mdadm with careful labeling, cheap, slow sata disks
1. folks mention that this particular setup with sillicon image chipsets top off at about 3 WDgreen 5.4k rpm drives’ worth of throughput; i don’t care about this, because this is for bulk storage and near-line backup. as long as data is reliably retrievable, it’s fine
2. folks mention that ncq should be disabled because a reset event on the PMP causes drives to lose data - need to test this before pushing to prod, and it’s abit of a dealbreaker since the whole idea of a port multiplier kinda hinges on being able to leverage ncq
3. enclosures are so overpriced :/ Would be awesome and worth it though if the disks were able to self-identify through mdadm whichever member disk failed so that remote-hands FRU capability can be had (on a bare-bones budget!)
4. RAID6 is a mindlessly easy and secure solution, but RAID 50 might be a sweeter price+performance spot.
Future:
1. think about another array for fast data storage using 2.5 in hybrid disks and a higher performance, caching sata/sas controller - basically the complete opposite of the above
2. leverage fuse-zfs for poor man’s direct-access* tiered storage (ok, i clearly have no idea what i’m talking about wrt tiered storage, except that I don’t at all want DMF-like tiered storage where a copy of the stuff lives on each tier). Ext4 + flashcache is fine too
3. How many SSDs do we need for a reasonable scratchdisk + access cache?
Also, on archiving: LTO4 is so damn expensive :( BD-R’s are getting cheap…
