[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

I need help testing new interface

From: Sergio Lopez
Subject: I need help testing new interface
Date: Thu, 13 May 2010 00:15:47 +0200


I've implemented a replacement for m_o_lock_request (called
m_o_sync_request) that helps to prevent thread storms in ext2fs when
synchronizing large pagers, by giving translators the ability to
throttle the amount of m_o_data_return request that are going to be

When using this interface, thread storms can still appear in some
situations, like when creating a large number of files sequentially
(when object's cache list is full, GNU Mach will start cleaning it, 
triggering a race of pageouts between vm_object_terminate and
synchronization thread of ext2fs) or when memory is almost exhausted.
These situations would be solved with a new cache system for memory
objects (and this is something we should start working on ASAP).

I also needed to change m_o_terminate interface to prevent RPCs to get
stuck (I've described this problem in a previous mail). We didn't
notice this previously because before m_o_sync_request, the only RPC
for memory object was m_o_get_attributes, which is not currently in use
by Hurd.

I need some help reviewing the patch (I consider this a preliminary
version, so there's no Changelog) and testing the stability of this
interface. Since it requires to rebuild GNU Mach, Hurd and glibc, I'll
try to publish patched packages to make easier testing it.


Attachment: gnumach-sync.diff
Description: Text Data

Attachment: hurd-sync.diff
Description: Text Data

reply via email to

[Prev in Thread] Current Thread [Next in Thread]