Mailing-List: contact user-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@cassandra.apache.org
Received-SPF: pass (athena.apache.org: domain of daemeonr@gmail.com designates
 209.85.223.171 as permitted sender)
MIME-Version: 1.0
In-Reply-To: <loom.20150305T012807-263@post.gmane.org>
References: <loom.20150305T012807-263@post.gmane.org>
From: daemeon reiydelle <daemeonr@gmail.com>
Date: Wed, 4 Mar 2015 17:29:04 -0800
Message-ID: 
 <CAOUOv0ECEPCd6QrfUDM68P3aZp+tQ_g96z7L5MD6_bbfL4jMKw@mail.gmail.com>
Subject: Re: Does it makes sense to split Gossip from Thrift network
To: user@cassandra.apache.org
Content-Type: multipart/alternative; boundary=20cf301b66a31ec00c0510808029

--20cf301b66a31ec00c0510808029
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

If your cluster is typical, your most critical resource is your network
bandwidth, if this is the case, I would not do this split you are
proposing. One issue with large MTU's is that they are often split at the
switch fabric. Switches are not generally known for having processors that
are idle, so this actually slows DOWN throughput to busy nodes.

I would also be concerned that in many cases the SAME nodes are supposed to
be identifiable as the SAME Gossip and Thrift related node. When you set
them up as different IP's I think you will will have additional issues.

What is the use case you are trying to solve with this configuration? If
Thrift is flooding your network port, then taking away bandwidth (assuming
you are bonding?) seems contraproductive. Unless you have not yet tried
bonding?


*.......*


*=E2=80=9CLife should not be a journey to the grave with the intention of a=
rriving
safely in apretty and well preserved body, but rather to skid in broadside
in a cloud of smoke,thoroughly used up, totally worn out, and loudly
proclaiming =E2=80=9CWow! What a Ride!=E2=80=9D - Hunter ThompsonDaemeon C.=
M. ReiydelleUSA
(+1) 415.501.0198London (+44) (0) 20 8144 9872*

On Wed, Mar 4, 2015 at 4:35 PM, Steffen Winther <cassandra.user@siimnet.dk>
wrote:

> Hi
>
> Wondering if if makes sense to split network for client traffic vs
> Gossip/Internode traffic (possible with larger MTU for storage traffic).
>
> So I tried this:
>
> - Gossip storage listener (port 700x) on one network
> - Thrift/CQL listeners (port 9160/9042) on another
>
> Only I find it a bit confusing to control exactly this,
> specially when trying to configure OpsCenter and agents.
>
> Are they any good doc pointers for splitting network?
>
> TIA
>
>

--20cf301b66a31ec00c0510808029
Content-Type: text/html; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div class=3D"gmail_default" style=3D"font-family:comic sa=
ns ms,sans-serif;color:rgb(7,55,99)">If your cluster is typical, your most =
critical resource is your network bandwidth, if this is the case, I would n=
ot do this split you are proposing. One issue with large MTU&#39;s is that =
they are often split at the switch fabric. Switches are not generally known=
 for having processors that are idle, so this actually slows DOWN throughpu=
t to busy nodes.<br><br></div><div class=3D"gmail_default" style=3D"font-fa=
mily:comic sans ms,sans-serif;color:rgb(7,55,99)">I would also be concerned=
 that in many cases the SAME nodes are supposed to be identifiable as the S=
AME Gossip and Thrift related node. When you set them up as different IP=
9;s I think you will will have additional issues.<br><br></div><div class=
=3D"gmail_default" style=3D"font-family:comic sans ms,sans-serif;color:rgb(=
7,55,99)">What is the use case you are trying to solve with this configurat=
ion? If Thrift is flooding your network port, then taking away bandwidth (a=
ssuming you are bonding?) seems contraproductive. Unless you have not yet t=
ried bonding?<br></div></div><div class=3D"gmail_extra"><br clear=3D"all"><=
div><div class=3D"gmail_signature"><div dir=3D"ltr"><div><div dir=3D"ltr"><=
span style=3D"color:rgb(56,118,29)"><span style=3D"background-color:rgb(255=
,255,255)"><b><span style=3D"font-family:comic sans ms,sans-serif"></span><=
/b></span></span><span style=3D"color:rgb(56,118,29)"><span style=3D"backgr=
ound-color:rgb(255,255,255)"><b><span style=3D"font-family:comic sans ms,sa=
ns-serif"><br>.......<br></span></b></span></span><span style=3D"color:rgb(=
56,118,29)"><span style=3D"background-color:rgb(255,255,255)"><b><span styl=
e=3D"font-family:comic sans ms,sans-serif"><strong>=E2=80=9CLife should not=
 be a journey to the grave with the intention of
 arriving safely in a<br>pretty and well preserved body, but rather to skid
 in broadside in a cloud of smoke,<br>thoroughly used up, totally worn out,
 and loudly proclaiming =E2=80=9CWow! What a Ride!=E2=80=9D</strong> <br>- =
Hunter Thompson<br><br>Daemeon C.M. Reiydelle<br>USA (+1) 415.501.0198<br>L=
ondon (+44) (0) 20 8144 9872</span></b></span></span><font size=3D"1"><i><b=
r></i></font></div></div></div></div></div>
<br><div class=3D"gmail_quote">On Wed, Mar 4, 2015 at 4:35 PM, Steffen Wint=
her <span dir=3D"ltr">&lt;<a href=3D"mailto:cassandra.user@siimnet.dk" targ=
et=3D"_blank">cassandra.user@siimnet.dk</a>&gt;</span> wrote:<br><blockquot=
e class=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;border-left:1px #ccc sol=
id;padding-left:1ex">Hi<br>
<br>
Wondering if if makes sense to split network for client traffic vs<br>
Gossip/Internode traffic (possible with larger MTU for storage traffic).<br=
>
<br>
So I tried this:<br>
<br>
- Gossip storage listener (port 700x) on one network<br>
- Thrift/CQL listeners (port 9160/9042) on another<br>
<br>
Only I find it a bit confusing to control exactly this,<br>
specially when trying to configure OpsCenter and agents.<br>
<br>
Are they any good doc pointers for splitting network?<br>
<br>
TIA<br>
<br>
</blockquote></div><br></div>

--20cf301b66a31ec00c0510808029--