[erlang-questions] Support for polling HDD in linux 4.4

[Max Lapshin]

Lukas, we can meet situation, when even SSD is too slow because it is
plugged via slow SATA cable that can do only 6 Gbit/s but we want 10 Gbit/s
or even 20


More often we have to handle load when there are 10 or 20 HDD  (JBOD, so no
raid, just 20 mounted devices) and we need to serve data from them.

If too many users want data from one disk, this disk blocks. Without any
explicit actions all disk requests to this disk are taking all threads from
async pool.

If we try to use   file:read_file at this moment to try to read from SSD,
we cannot read it, because all threads from thread pool are blocked on slow
HDD.

This is why we maintain our own internal queue with timeout per each device
and try to send 10-40 simultaneous requests to each device.

It is almost impossible to do with LVM or RAID and completely impossible to
do with hardware RAID (but hardware raid is not about speed), but such
approach is possible with separate HDD.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20160424/67218582/attachment.htm>