On 11/06/2014 07:57 PM, John Andersen wrote:
On 11/06/2014 04:51 PM, Carlos E. R. wrote:
It uses only one thread and one core, even when having several large disks and dozens of free cpu cores. Thus it can not parallelize writes.
Probably in recognition of the fact that disks have exactly one controller, on on board processor, one arm.
No matter how much sand you have and how big a hammer you use, you can only pound so much sand down a single rabbit hole at any given time.
PDP-11 and early Vaxen had a similar problem. The dis controller could support up to 8 spindles. Good, you think. 4K file system, 512 byte sectors, 8 drives, PARALLELISM! No! Could only transfer to one spindle/arm at a time. best effort, of you hand managed the control queue, and had a controller with dedicated DMA that could fetch the control blocks that you set up and chained was to have seek1 seek2 seek3 seek4 transfer1 seek5 transfer2 seek6 transfer3 seek7 transfer4 seek8 transfer5 wait transfer6 wait transfer7 wait transfer8 Of course you _could_ get much better performance by installing a second controller. Heck, when metricated it tuned out that just having two drives was usually faster. Particularly on the PDP-11 which was a roll-in-roll-out type swap architecture, no virtual memory & paging. Having a drive dedicated to swap was very efficacious. Some time in the not to distant future I'm going to experiment with have ROOT as striped BtrFS on two SATA drives. In the long run I think I'll settle for mirroring, but along the way ... -- A: Yes. > Q: Are you sure? >> A: Because it reverses the logical flow of conversation. >>> Q: Why is top posting frowned upon? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org