pcp
[Top] [All Lists]

Re: pmlogger -u questions

To: Nathan Scott <nathans@xxxxxxxxxx>
Subject: Re: pmlogger -u questions
From: "Frank Ch. Eigler" <fche@xxxxxxxxxx>
Date: Mon, 14 Apr 2014 20:29:52 -0400
Cc: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>, pcp@xxxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <155006091.5545657.1397518977813.JavaMail.zimbra@xxxxxxxxxx>
References: <01e901cf56df$4ce97de0$e6bc79a0$@internode.on.net> <1665962954.4723287.1397437104781.JavaMail.zimbra@xxxxxxxxxx> <534B4330.1060008@xxxxxxxxxxxxxxxx> <y0meh104nvl.fsf@xxxxxxxx> <534C4FF4.5000304@xxxxxxxxxxxxxxxx> <20140414212551.GK14108@xxxxxxxxxx> <534C6531.6050502@xxxxxxxxxxxxxxxx> <155006091.5545657.1397518977813.JavaMail.zimbra@xxxxxxxxxx>
User-agent: Mutt/1.4.2.2i
Hi -

On Mon, Apr 14, 2014 at 07:42:57PM -0400, Nathan Scott wrote:
> [...]
> I concur with Frank here - definitely warrants some deep cost analysis -
> it sounds risky WRT introducing performance regression.  

I wouldn't over-worry about it.  The PMCD PDU's are already being
relayed to volume files in one chunk, so that's great.  Getting the
trailer bytes in there would be dandy.  Beyond that, we'd be actually
saving some effort in that we'd eliminate one copy (by not going into
stdio buffers).

> What is the I/O size we generally submit from pmlogger data writes,
> OOC?  [...]

Ken reported seeing alternating 4/8K write(2) syscalls coming, which
makes sense for 12K-ish log record batches going into stdio.


- FChE

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