aufs in Squid-2.3/2.4.DEVEL is quite buggy and behaving strangely.
I am currently trying to do a major cleanup of aufs before 2.4 gets
released. See http://squid.sourceforge.net/projects.html#asyncio
Yes. As you note the MAGIC1 in the distributed sources does not make
much sense. A more sensible MAGIC1 is
#define MAGIC1 (NUMTHREADS*Config.cacheSwap.n_configured/2)
IIRC similar problems apply to MAGIC2.
MAGIC1/MAGIC2 is only one of the problems. There are several other
problems. Some old, some introduced in Squid-2.3 when the code got
separated from the rest of the code.
-- Henrik Nordstrom Squid hacker cachemail@procache.com wrote: > > Dear squid-dev > I saw the following line in the storeAufsOpen function in the > ./src/store_io_asyncufs.c file. Using of this is rational, because > it prevents the squid from being overload. > > " while (aioQueueSize() > MAGIC1) > return NULL;" > > MAGIC1 is defined in the upper part as: > #define MAGIC1 (NUMTHREADS/Config.cacheSwap.n_configured/2) > > I do not know what's the reason for defining the MAGIC1 in this way? > Please let me know more informatiom in this regard. > > Best Regards, > V.FkReceived on Tue Oct 31 2000 - 11:47:27 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:12:53 MST