Shared Synchronization Objects

Jeff V. Merkey (jmerkey@timpanogas.com)
Fri, 14 May 1999 09:56:45 -0600


This is a multi-part message in MIME format.

------=_NextPart_000_000E_01BE9DF0.132C3A10
Content-Type: text/plain;
charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable

Linux Community,

I've seen discussion of shared sync objects for SMP (reader/writer =
locks) from some folks. I have not seen anything out there for them in =
the kernel sources. We really need this in Linux. In particular, they =
would make file systems ***SMOKE*** from a performance standpoint. =
Comms (neworking) subsystems usually are ok with simple mutexes and spin =
locks, but file systems and databases are different beasts. I've =
implemented them in two different SMP kernels. It sounds like several =
folks are working on implementations of them. I'd like to see them get =
rolled into the kernel -- they are desperately needed.

Standards are good. the UI (Unix International) APIs are some of the =
better for defining SMP synchronization objects (mutex_lock, =
mutex_unlock, rd_rwlock, etc.) becuase they are standardized, easy to =
implement, and well known. They also can be implemented in-kernel. If =
someone else doesn't take stab at putting them in, we may try to, if =
folks think they would be a good idea. We really need a shared sync =
object in the kernel for file system work (shared FATs, Dir Tables, =
etc.) they increase SMP parallelism and scaling if implemented correctly =
very significantly. =20

If other folks are interested in this, we could take a shot at putting =
them in, but we aren't religious about it, and would certainly welcome =
other folks to try as well -- it will help all of us.

Please advise.

Jeff

------=_NextPart_000_000E_01BE9DF0.132C3A10
Content-Type: text/html;
charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable

<!DOCTYPE HTML PUBLIC "-//W3C//DTD W3 HTML//EN">

 
Linux Community,
 
I've seen discussion of shared sync = objects for=20 SMP (reader/writer locks) from some folks.  I have not seen = anything out=20 there for them in the kernel sources.  We really need this in = Linux. =20 In particular, they would make file systems ***SMOKE*** from a = performance=20 standpoint.  Comms (neworking) subsystems usually are ok with = simple=20 mutexes and spin locks, but file systems and databases are different=20 beasts.  I've implemented them in two different SMP kernels.  = It=20 sounds like several folks are working on implementations of them.  = I'd like=20 to see them get rolled into the kernel -- they are desperately=20 needed.
 
Standards are good.  the UI = (Unix=20 International) APIs are some of the better for defining SMP = synchronization=20 objects (mutex_lock, mutex_unlock, rd_rwlock, etc.)  becuase they = are=20 standardized, easy to implement, and well known.  They also can be=20 implemented in-kernel.  If someone else doesn't take stab at = putting them=20 in, we may try to, if folks think they would be a good idea.  We = really=20 need a shared sync object in the kernel for file system work (shared = FATs, Dir=20 Tables, etc.) they increase SMP parallelism and scaling if implemented = correctly=20 very significantly. 
 
If other folks are interested in = this, we could=20 take a shot at putting them in, but we aren't religious about it, and = would=20 certainly welcome other folks to try as well -- it will help all of=20 us.
 
Please advise.
 
Jeff
 
 
------=_NextPart_000_000E_01BE9DF0.132C3A10-- - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/