Bug 60036
Summary: | sard-patch time statistics in /proc/partitions wrap too soon | ||
---|---|---|---|
Product: | [Retired] Red Hat Linux | Reporter: | Need Real Name <br6zm9h001> |
Component: | kernel | Assignee: | Stephen Tweedie <sct> |
Status: | CLOSED CURRENTRELEASE | QA Contact: | Brian Brock <bbrock> |
Severity: | low | Docs Contact: | |
Priority: | medium | ||
Version: | 7.2 | ||
Target Milestone: | --- | ||
Target Release: | --- | ||
Hardware: | All | ||
OS: | Linux | ||
Whiteboard: | |||
Fixed In Version: | Doc Type: | Bug Fix | |
Doc Text: | Story Points: | --- | |
Clone Of: | Environment: | ||
Last Closed: | 2003-06-08 01:30:11 UTC | Type: | --- |
Regression: | --- | Mount Type: | --- |
Documentation: | --- | CRM: | |
Verified Versions: | Category: | --- | |
oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
Cloudforms Team: | --- | Target Upstream Version: | |
Embargoed: |
Description
Need Real Name
2002-02-19 07:48:18 UTC
That's not a bug. The sard stats were never designed to track total IO, but rather to support interval IO tracking. "iostat" will take two successive snapshots of these stats and will display the ongoing IO statistics from them by taking the difference, and that should still work even when the counters wrap. btw, most of the hz_to_std routines have the same overflow property. I'll mull this over --- it may well be worth cleaning this up somewhat. iostat works fine when counters wrap at 2^32, but these ones wrap at 2^32 / 100. Here's an output of iostat for an interval where aveq wrapped over, resulting in bogus avgqu-sz: Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s avgrq-sz avgqu-sz await svctm %util hda 27.96 4.19 13.78 4.79 318.40 69.63 20.89 70921.74 2921.21 43.84 8.14 Yes, I realised that just after the previous reply. I'm currently testing a kernel with #define MSEC(x) (((x) / HZ) * 1000 + ((x) % HZ) * 1000 / HZ) defined unconditionally --- the compiler seems to create decent code for that. (gcc has a number of very nice optimisations for such expressions.) Fix checked in. |