<div dir="ltr"><div>My apologies Randal, I forgot to forward along Eric's answers to your questions to the list.<br></div>The presentation's slides are available on the CALUG website.<br><div><br><div><div class="gmail_quote">
---------- Forwarded message ----------<br>From: <b class="gmail_sendername">Eric Sproul</b> <span dir="ltr"><<a href="mailto:esproul@omniti.com" target="_blank">esproul@omniti.com</a>></span><br>
Date: Thu, Nov 7, 2013 at 1:41 PM<br>Subject: Re: [CALUG] November 13 Meeting Announcement - ZFS Overview<br>To: Keith <<a href="mailto:keith@keithf4.com" target="_blank">keith@keithf4.com</a>><br><br><br>I'll answer the questions as best I can and let you forward them back<br>
to the list.<br>
<br>
On Thu, Nov 7, 2013 at 12:13 PM, Keith <<a href="mailto:keith@keithf4.com" target="_blank">keith@keithf4.com</a>> wrote:<br>
><br>
><br>
> ---------- Forwarded message ----------<br>
> From: Randal T. Rioux <<a href="mailto:randy@procyonlabs.com" target="_blank">randy@procyonlabs.com</a>><br>
> Date: Wed, Nov 6, 2013 at 12:08 AM<br>
> Subject: Re: [CALUG] November 13 Meeting Announcement - ZFS Overview<br>
> To: <a href="mailto:calug@unknownlamer.org" target="_blank">calug@unknownlamer.org</a><br>
<br>
<br>
> One thing that attracted me is the snapshot capability, and that I can<br>
> set gzip-7 at the filesystem level (avg 1.18x compression right now!)<br>
> and transfer are actually *faster* with compression on :-)<br>
<br>
This is true, because most file servers have CPU cycles to spare, and<br>
trading extra CPU for reduced IOPS is almost always a win. I'd note<br>
that with mostly uncompressable data (1.18x would qualify), the<br>
default compression algorithm (lzjb) will likely be even faster than<br>
gzip, especially at level 7. The more recent addition of lz4 should<br>
be faster still, especially on uncompressable data.<br>
<br>
I'll be covering compression options in the talk.<br>
<br>
> One quick one now if you don't mind before the talk: what is the best<br>
> way to schedule a snapshot regimen for 1, 3, 7 and 30 days including<br>
> destroying old ones? I see scripts and stuff on the Web, but they all<br>
> look different and seem to over complicate things.<br>
<br>
"Best way" invites lots of opinions, so I'll simply say that snapshots<br>
may be created at will and destroyed without respect to sequence. So<br>
you can take snapshots at some minimum interval, then later on make a<br>
decision about which ones to keep or delete.<br>
<br>
><br>
> Tangent: can you tell us more about OmniOS (online info is scarce) and<br>
> whether Circonus is purely commercial (appears so)?<br>
<br>
OmniOS is a distribution of illumos (<a href="http://illumos.org" target="_blank">http://illumos.org</a>) which is an<br>
open-source fork of the former OpenSolaris codebase. It is a<br>
minimalist distribution aimed at traditional enterprise server<br>
deployments. The mission statement is here:<br>
<a href="http://omnios.omniti.com/wiki.php/OmniOSJeOS" target="_blank">http://omnios.omniti.com/wiki.php/OmniOSJeOS</a> and contains a link to<br>
Theo's LISA'12 talk, "OmniOS Motivation and Design".<br>
<br>
Circonus is indeed a commercial product, but the metrics collection<br>
engine that it uses is the open-source Reconnoiter project:<br>
<a href="https://github.com/circonus-labs/reconnoiter" target="_blank">https://github.com/circonus-labs/reconnoiter</a><br>
<span><font color="#888888"><br>
Eric<br>
</font></span></div><br></div></div></div>