SYSTEM ARCHITECTURE COUNCIL
Platform Software ARC
---------------------------------
PSARC Regular Meeting time: Wednesdays 10:00-1:00pm in MPK17-3507.
09-02-2009 MEETING MINUTES
============================================================================
Send CORRECTIONS, additions, deletions to psarc-coord at sun.com.
Minutes are archived in sac.Eng:/sac/export/sac/Minutes/PSARC.
Co-Chair(s):
Garrett D'Amore: Yes
Tim Marsland: no
ATTENDEES - Members: (6 active members)
Kais Belgaied: Yes
Mark Carlson: no
Richard Matthews: Out
Darren Moffat: no (on sabbatical)
Sebastien Roy: Yes
Glenn Skinner: Yes
Bill Sommerfeld: no (on sabbatical)
Gary Winiger: no (on sabbatical)
STAFF -
Asa Romberger (PM): Yes
ATTENDEES - Interns:
Frank Che no
James Falkner: no (on sabbatical)
Daniel Hain: Yes
Michael Haines: no
Alan Hargreaves: no
Phil Harman: no
Wyllys Ingersoll: no
Darren Reed: no
Dean Roehrich no
Ienup Sung: no
Phi Tran Yes
Brian Utterback: no
James Walker Yes
Suha Peddada Yes
Mark Martin Yes (external)
Don Cragun Yes (external)
Guests:
-- GUESTS --
Rich Brown
Tim Haley
Shrikrishna Khare
Vladimir Kotal
Nicolas Droux
Not all names are captured. Please send email to Asa.Romberger at Sun.com, if
you attended the meeting and your name is missing from the list.
---------------------------------------------------------------------------
MEETING SUMMARY:
================
AGENDA
09/02/2009
10:00-10:10 Open ARC Business (use open dial in above)
10:10-10:55 Open Inception: dlstat and flowstat (2009/364)
Submitter: Shrikrishna Khare
Owner: Kais Belgaied
Exposure: open
---------------------------------------------------------------------------
Case Anchors: <br>
<A HREF="#case1">dlstat and flowstat (2009/364)</A> <br>
===========================================================================
Fast Tracks:
============
Case (Timeout) Exposure Title
2009/396 (09/09/09) open Tickless Kernel Architecture / lbolt
decoupling
let it run
2009/444 (08/24/09) open daemon() in libc
approved
2009/449 (09/02/09) open sshd PreUserAuthHook
approved
2009/455 (08/28/09) open Environment Modules
more time 09/09/09
2009/459 (09/01/09) open Logical Domains Agents on Solaris
approved
2009/461 (09/02/09) open Transport Layer Retries (TLR) Support
approved
2009/464 (09/02/09) open Offline attribute
approved
2009/465 (09/02/09) open COMSTAR ALUA active/standby support
more time 09/05/09
2009/468 (09/04/09) open nfs4_fid()
approved
Next Meeting:
=============
09/09/2009
10:00-10:10 Open ARC Business (use open dial in above)
10:15-10:25 Closed ARC Business (use closed dial in above)
10:25-11:10 Closed Inception: Public GLDv3 Interfaces
Submitter: Nicolas Droux
Owner: Kais Belgaied
Exposure: closed
===========================================================================
2009/364
Name: dlstat and flowsat
Submitter: Shrikrishna Khare
Owner: Kais Belgaied
Status: inception held 06/24/2009
Exposure: open
SUMMARY
=======
We propose to provide new commands to monitor dynamic network traffic
statistics viz. dlstat(1M) and flowstat(1M). This is part of the effort
to gain better visibility into network traffic in the light of crossbow
features like virtual NICs, interrupt vs. polling modes etc. This would
greatly assist network performance analysis. It is also aimed at
segregating link/flow configuration interface from the interface for
querying for dynamic network statistics.
ISSUES
======
seb-1 I don't see an obvious benefit of splitting up dlstat and
flowstat. Why would dlstat instead not have a mode whereby it
prints flow statistics given that flows can't accumulate
statistics unless they're associated with a datalink. IMO,
introducing two closely-interrelated commands with almost
identical command-lines and overall operation is not
necessary. I don't feel religious about it, but would like to
hear the opinions on this.
seb-2 There is seemingly nothing differentiating the first two
dlstat usages in the man page. Is it simply the case that -t,
-r, and -o are each mutually exclusive? If so, there's
probably a way to represent that explicitly in the usage line.
seb-3 The use of "-o [+|-]field,..." is confusing to me, and is
unlike dladm and flowadm's use of -o to select output fields.
A simpler model would be to have a default set of fields, and
if -o is used, the caller must specify the complete set of
fields needed. This is much less ambiguous. The other
advantage to this is that the user's output, when -o is used,
will be constant from release to release even if default
fields change.
seb-4 The "show" subcommand is contrary to the existing convention
of using <action>-<object> for subcommands. On a related
note, it's unclear that you need any subcommands for these
commands given that the whole purpose of the commands is to
show statistics.
seb-5 What is the intended purpose of the "reset" subcommands? I
know _what_ they do, but can't think of a case where this
would be useful and not harmful (e.g. by confusing management
monitoring tools that expect counters to generally flow
uphill). If this has benefitial use, it should probably be
surounded by appropriate warnings in documentation.
ged-1 Why aren't the underlying statistics available via kstats?
(We already have kstats for reporting general purpose stats.)
ged-2 What is the underlying mechanism used between the commands and
the MAC layer?
ged-3 Will these statistics be available via some form of library
to C programs (libdladm)?
ged-4 Man page nit: the language in dladm referencing show-link
still shows -s and -i in the usage.
dwc-1 Building on top of seb-3:
1. Having an option with more than 1 option-argument is bad.
2. Having an option with an optional option-argument is worse.
3. Having an option specifying multiple fields be required to
be presented as a single option (as opposed to multiple
occurrences of -o field) is onerous, especially when you
allow adding or deleting output fields from a base default.
Why not allow
[-o field[,...]]...
or, if you insist on adding or deleting from a default,
[-o [+|-]field]...
or
[-o [+|-]field[,[+|-]field]...]...
dwc-2 Why are we removing the descriptions of several options in dladm
and flowadm when they have committed interface stabilities.
jdc-01 I suggest getting rid of "reset" altogether. Besides being
fundamentally incompatible with SNMP instrumentation, "reset" just
isn't necessary or complete as long as you have decent delta-
calculating tools. (And I'd really rather have a good way of
computing deltas -- especially as a non-privileged user -- than having
an inaccessible way to nuke kernel counters.)
Commitment issues:
seb-6 The case isn't explicit about the changes to dladm. It uses
language like "migrate" to describe where that the
functionality will "move" to dlstat, but the dladm man page
still has "-s" and friends in the various dladm subcommand
synopses. This is confusing. Are the various "-s" options in
dladm being removed? Please clarify.
dwc-3 dladm(1M) and flowadm(1M)
dwc-2 has not been addressed from the inception review.
Completely removing the description of committed obsolete
features with just text saying that the options are obsolete
seems extreme. The old description should remain with a new
leading paragraph saying that this usage is now obsolete,
XXX(1M) is the preferred way to get this functionality.
dwc-97 dlstat(1M), Example 9
The fields in the example output don't match the list of fields
requested in the -o option argument.
dwc-98 flowstat(1M), Example 3
I believe the command line for this description should be:
flowstat -t -i5
instead of:
flowstat -t 5
dwc-99 flowstat(1M), description of -s and -t options
You should explicitly state that "hh" is using 24-hour clock
notation.
THE NEXT STEP
=============
Approved
Garrett D'Amore
Kais Belgaied
Sebastien Roy
Glenn Skinner
