Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 27 Feb 2015 02:44:13 +0000 (UTC)
From:      "Kenneth D. Merry" <ken@FreeBSD.org>
To:        src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org
Subject:   svn commit: r279346 - head/lib/libdevstat
Message-ID:  <201502270244.t1R2iDiI080234@svn.freebsd.org>

next in thread | raw e-mail | index | archive | help
Author: ken
Date: Fri Feb 27 02:44:12 2015
New Revision: 279346
URL: https://svnweb.freebsd.org/changeset/base/279346

Log:
  Remove an obsolete comment in devstat(3) about the accuracy of the
  milliseconds per transaction (DSM_MS_PER_TRANSACTION) calculation.
  
  The comment was accurate many years ago when the kernel didn't
  record I/O times on a per-I/O basis, but now that we do collect
  that information in most areas, it isn't correct.
  
  The milliseconds per transaction values are correct, assuming the
  I/O duration has been recorded.
  
  Sponsored by:	Spectra Logic
  MFC after:	3 days

Modified:
  head/lib/libdevstat/devstat.c

Modified: head/lib/libdevstat/devstat.c
==============================================================================
--- head/lib/libdevstat/devstat.c	Fri Feb 27 02:21:52 2015	(r279345)
+++ head/lib/libdevstat/devstat.c	Fri Feb 27 02:44:12 2015	(r279346)
@@ -1487,22 +1487,9 @@ devstat_compute_statistics(struct devsta
 				*destld = 0.0;
 			break;
 		/*
-		 * This calculation is somewhat bogus.  It simply divides
-		 * the elapsed time by the total number of transactions
-		 * completed.  While that does give the caller a good
-		 * picture of the average rate of transaction completion,
-		 * it doesn't necessarily give the caller a good view of
-		 * how long transactions took to complete on average.
-		 * Those two numbers will be different for a device that
-		 * can handle more than one transaction at a time.  e.g.
-		 * SCSI disks doing tagged queueing.
-		 *
-		 * The only way to accurately determine the real average
-		 * time per transaction would be to compute and store the
-		 * time on a per-transaction basis.  That currently isn't
-		 * done in the kernel, and would only be desireable if it
-		 * could be implemented in a somewhat non-intrusive and high
-		 * performance way.
+		 * Some devstat callers update the duration and some don't.
+		 * So this will only be accurate if they provide the
+		 * duration. 
 		 */
 		case DSM_MS_PER_TRANSACTION:
 			if (totaltransfers > 0) {
@@ -1512,11 +1499,6 @@ devstat_compute_statistics(struct devsta
 			} else
 				*destld = 0.0;
 			break;
-		/*
-		 * As above, these next two really only give the average
-		 * rate of completion for read and write transactions, not
-		 * the average time the transaction took to complete.
-		 */
 		case DSM_MS_PER_TRANSACTION_READ:
 			if (totaltransfersread > 0) {
 				*destld = totaldurationread;



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201502270244.t1R2iDiI080234>