| Title: | "ASK THE WIZARDS" |
| Moderator: | QUARK::LIONEL |
| Created: | Mon Oct 30 1995 |
| Last Modified: | Mon May 12 1997 |
| Last Successful Update: | Fri Jun 06 1997 |
| Number of topics: | 1857 |
| Total number of notes: | 3728 |
Return-Path: "VMS001::WWW"@vms001.das-x.dec.com Received: by vmsmkt.zko.dec.com (UCX V4.1-12, OpenVMS V6.2 VAX); Thu, 27 Feb 1997 08:39:45 -0500 Received: from vms001 by mail12.digital.com (8.7.5/UNX 1.5/1.0/WV) id IAA30534; Thu, 27 Feb 1997 08:32:09 -0500 (EST) Date: Thu, 27 Feb 1997 08:33:33 -0500 Message-Id: <[email protected]> From: "VMS001::WWW"@vms001.das-x.dec.com (27-Feb-1997 0833) To: [email protected], [email protected], [email protected] Subject: Ask the Wizard: '[email protected]' X-VMS-To: [email protected] Remote Host: (null) Browser Type: Mozilla/3.0b4Gold (X11; I; OpenVMS V7.0 AlphaStation 200 4/233) Remote Info: <null> Name: jeff byrkit Email Address: [email protected] CPU Architecture: Alpha Version: v 7.0 Questions: we are considering purchasing the as 4100 -300e system with 4 CPU's. Our dbms most likely is not SMP. If we run on this box, we we only be utilizing the other three processors if we have SMP programs? will all of our interactive processes be on the same CPU? or can we load-balance between the 4 processors? Great service!
| T.R | Title | User | Personal Name | Date | Lines |
|---|---|---|---|---|---|
| 1639.1 | OpenVMS SMP | XDELTA::HOFFMAN | Steve, OpenVMS Engineering | Thu Feb 27 1997 15:39 | 30 |
The `S' in `SMP' is `Symmetric' -- you need do nothing more than run four `computable' processes to use those four processors. Unless specifically constrained by the programmer or by the system manager, any computable process can (and will) be scheduled on the next available processor in an OpenVMS symmetric multi-processor (SMP) system. Typically the standard OpenVMS scheduler is the best choice for handling this process scheduling... But there are ways to alter the behaviour of the scheduling: With `kernel threading' enabled (OpenVMS Alpha V7.0 and later, only), a single `thread-aware' process can be scheduled across multiple processors fully in parallel; threads of execution run entirely in parallel. For information on this, see the DECthreads manual. One can constrain which processors can be used using the (recently added) command SET PROCESS/AFFINITY, and one can customize the existing process scheduling scheme using a user-written class scheduler (for an example of this, see sys$examples:class.c). The pendantically-inclined will point out that a few specific low-level tasks -- such as the system clock or the processing of an I/O *completion* operation -- must be performed on a specific processor in an OpenVMS SMP system. | |||||