Red Hat Bugzilla – Bug 475548
[LTC 5.4 FEAT] FCP - Performance data collection (blktrace) 
Last modified: 2009-06-20 04:44:15 EDT
Emily J. Ratliff <firstname.lastname@example.org> -
1. Feature Overview:
Feature Id: 
a. Name of Feature: FCP - Performance data collection (blktrace)
b. Feature Description
The item comprises:
a) facilities which provide performance relevant data (see below) at any time
b) facilities which collect the data provided by a), then associate it with a time stamp and and
store it for future processing
- these facilities are configurable in regard of the type and amount of data collected for the
purpose of reducing the amount of temporarily stored data.
- when data is taken as samples, the sampling interval is configurable
- collected data can be filtered by e.g masks for valid stack components (see below) before
data is stored for further processing
Data collected comprises:
a) Performance relevant data about stack components as Linux devices, SCSI LUNs and HBA and storage
b) Per stack component: current values of relevant measurements as throughput, utilization and
other applicable measurements
c) Statistical aggregations (min, max average, histogram) of data associated with I/O requests as
size, latency per component and in total.
Pick up latest version from blktrace package and if IBM contribution, not yet included, apply it.
RHEL 6.0 request: https://bugzilla.redhat.com/show_bug.cgi?id=464179
2. Feature Details:
Arch Specificity: Both
Affects Installer: Yes
Delivery Mechanism: Backport
Request Type: Package - Feature from IBM
d. Upstream Acceptance: Submitted
Sponsor Priority 1
f. Severity: High
IBM Confidential: no
Code Contribution: IBM code
g. Component Version Target: > blktrace 0.99.3
3. Business Case
This item will make possible to have a better understanding of the SCSI activities and performance
of the customer system, being able afterward to better solve FCP issues regarding customer's SCSI
4. Primary contact at Red Hat:
5. Primary contacts at Partner:
Project Management Contact:
Hans-Georg Markgraf, email@example.com, Boeblingen 49-7031-16-3978
Gonzalo Muelas Serrano, firstname.lastname@example.org
Thomas Schwarz, email@example.com
What would be the exact steps to satisfy this request? Is it just a matter of including blktrace or are there other kernel or userspace requirements?
The code for this feature is upstream accepted is with blktrace 1.0.0
This feature has the coresponding feature requests:
201591 FCP - Performance Data collection & analysis (userspace) LTC 50340 RHBZ 475333
201590 FCP - Performance Data collection (kernel) Submitted LTC 50339 RHBZ 475334
The blktrace version to be included is the upstream version plus
the following two commits with bug fixes that are required for this feature, and will be upstream integrated within the next blktrace version
(git commit 61165962f86cb2de0b163f6083bff5649f692e91 and b3ee499dd143aeaed311bf3bc6d14871168b6657).
Attention, the maintainer has noch changed the version number of blktrace.
The existing kernel config should already support the blktrace tool, FWIW.
Assigning to Eric
For commit b3ee499dd143aeaed311bf3bc6d14871168b6657 as noted in comment #3 - "__BLK_TA_ABORT" and blk_abort_queue() are not currently present in the rhel5 kernel, so I don't think this fix is needed in the RHEL5 blktrace, is this correct?
It is absolutely necessary that the kernel API and blktrace user space match.
And we need to make sure that the blktrace API doesn't change (apart from additions).
So, if you leave out ABORT, we still need to make sure that the id for the ABORT is reserved and that it is not used by DRV_DATA. We don't want to reshuffle those ids later on, when ABORT gets added eventually.
+ __BLK_TA_ABORT, /* request aborted */
__BLK_TA_DRV_DATA, /* binary driver data */
It would be feasible to explicitely list the assigned id for DRV_DATA in that enum.
My preferred solution would be to put in ABORT, at least in the header file, because this would be closest to the upstream version.
Sure, that makes sense, thanks.
*** This bug has been marked as a duplicate of bug 472898 ***
Note, these 2 patches are now committed & built in blktrace-1.0.0-3.el5.