Re:A Tour Inside CloudFlare's Latest Generation Servers
by lotyrin on Monday July 22, 2013 @03:18PM
(#6085876)
No, it's just that their workload is not very well served by any RAID levels: many small files of which it's okay to completely lose a disk's worth and are very randomly and unevenly accessed.
They actually only wanted load balancing and I'm sure that their purpose built solution does a better job of being balanced while avoiding increased risk from striping or performance loss from mirroring or parity (I'm curious what level(s) they were using). Though, cutting out the RAID layer when they didn't need it does save them a trip through the controller, which is more important these days when compared to SSD "seek" times.
Re:A Tour Inside CloudFlare's Latest Generation Servers
by eastdakota on Monday July 22, 2013 @03:22PM
(#6085903)
Yup.
Wow 50%! Is this because raid controller performance hasn't kept up with the evolution from spinning disks to SSD's, or have raid controllers always had that much overhead?