Re: [RFC] cache architecture

From: Alex Rousskov <rousskov_at_measurement-factory.com>
Date: Tue, 24 Jan 2012 12:34:22 -0700

On 01/24/2012 03:02 AM, Pieter De Wit wrote:

>> I wonder about priority queues every now and then. It is an
>> interesting idea. The I/O is currently done with pluggable modules for
>> various forms. DiskThreads and AIO sort of do this but are FIFO queued
>> in N parallel queues. Prioritised queues could be an interesting
>> additional DiskIO module.

> Hard to implement given the current "leg work" is already done ? How
> well does the current version of squid handle multicores and can this
> take advantage of cores ?

Rock store can be configured to have a core-dedicated disker process for
each physical disk, with IPC queues connecting core-dedicated workers
and diskers. Rock queues do have some prioritization logic, but more
work is needed in that direction.

IMHO, the best options are not based on queue lengths and similar
easy-to-code parameters meaningless to an average cache administrator.
They should be based on response time, hit/miss bias and similar
concepts that cache administrators and folks they report to can understand.

However, discussing priority queues and options to control them is
probably outside this thread scope because the best choices and options
would depend on the storage module. It is also a very complex topic,
with some decisions/optimizations being OS- and even filesystem-specific.

HTH,

Alex.
Received on Tue Jan 24 2012 - 19:36:00 MST

This archive was generated by hypermail 2.2.0 : Wed Jan 25 2012 - 12:00:11 MST