pcp
[Top] [All Lists]

Re: pmdagfs2: Updates from continued testing and changes for future upda

To: Nathan Scott <nathans@xxxxxxxxxx>
Subject: Re: pmdagfs2: Updates from continued testing and changes for future updates
From: Paul Evans <pevans@xxxxxxxxxx>
Date: Thu, 10 Oct 2013 10:12:04 +0100
Cc: PCP Mailing List <pcp@xxxxxxxxxxx>
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <803653382.4750679.1381361383830.JavaMail.root@xxxxxxxxxx>
References: <525590DB.3080007@xxxxxxxxxx> <803653382.4750679.1381361383830.JavaMail.root@xxxxxxxxxx>
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20130625 Thunderbird/17.0.7
Hi Nathan,

On 10/10/2013 12:29 AM, Nathan Scott wrote:
FYI - 3.8.5 is pending, a day or two more if you're keen for it in this
release, else there's always next time.  Also, there's a handful of
Coverity-scan-reported issues in pmdagfs2 code if you want to take a
look, else I'll pick 'em up on my way through the rest of the code base.
I'll take a look see through these this morning and sort through them.
Please let me know of any issues and feedback is welcome :).
Couple of small things on review:

- GLOCK_ARRAY_CAPACITY setting seems a little coarse.  From some quick
   calcs, I think thats going to allocate an ~7-8MB array and then bump
   it up by that same amount each time its consumed.  Is 100000 glocks
   expected to be average/normal?  Seems alot ... an alternative scheme
   (which is still nice 'n simple & that we use elsewhere too) would be
   to start smaller (say 256 glocks), and double the buffer alloc size
   on each increment - this tries to balance quickly reaching the point
   where the buffer is well-sized for steady state, with consuming less
   memory at small glock counts.

- gfs2_ftrace.[ch] - I woulda just named em ftrace.[ch] since we're in
   the gfs2 directory here.  *shrug*
On a cluster under high load it is not uncommon for there to be over a million glocks at one time, understandably not all of them will be looked at during one refresh pass but I agree 100,000 is a little excessive and shall lower the value to a more acceptable range. Will also go about changing the name of gfs2_ftrace.[ch].

cheers.

ps: please let me know if there's more planned for within a day or two,
so I can ensure everything gets merged in time ... thanks!

--
Nathan
Cheers,

Paul

<Prev in Thread] Current Thread [Next in Thread>