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: local policy)
DomainKey-Signature: a=rsa-sha1; c=nofws; d=thelastpickle.com; h=to:from
	:subject:date:message-id:content-type:mime-version:in-reply-to;
	 q=dns; s=thelastpickle.com; b=0y7c3As9Ha4lI0mYpwauTh9ZuIEALXuuY
	u4EZZ4xWnP9XbOg1UUjJIBFRg7WPsB3UvOi/3o6R3tzPZiCZLNFvqTrtxvozZnnn
	HejKE9jc2Dnt8h/Q/cLcDWVQV4uF9VmpHsMNbfCLKXuNWY8jatnVwRUMRjqbNDg7
	HtlhhkqCOM=
To: user@cassandra.apache.org
From: Aaron Morton <aaron@thelastpickle.com>
Subject: Re: Smart Client Routing
Date: Wed, 17 Nov 2010 03:17:28 GMT
Message-id: <847de2cf-924d-491e-fa0b-d63e01ba6c30@me.com>
Content-Type: multipart/alternative;
 boundary=Apple-Webmail-42--6c0a9470-6bd9-ca28-e82a-c9ce23d0f40f
MIME-Version: 1.0
In-Reply-To: <423288.88728.qm@web51107.mail.re2.yahoo.com>


--Apple-Webmail-42--6c0a9470-6bd9-ca28-e82a-c9ce23d0f40f
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain;
    charset=ISO-8859-1;
    format=flowed

No need to worry.=A0=0A=0AI run REST requests through Varnish box > nginx =
/ Tornaod / Python box > Cassandra cluster and can get requests in and out=
 of the stack in a couple of milliseconds. Using some old workstation HW a=
nd not paying much attention to tuning.=A0=0A=0ABuild it like a normal sys=
tem and separate out the parts, if / when you have problems then you can l=
ook at tuning the cassandra cluster or other parts of the stack. There are=
 normally a number of other issues to deal with before network IO.=0A=0AHo=
pe that helps.=A0=0AAaron=0A=0A=0AOn 17 Nov, 2010,at 03:31 PM, E S <tr1skl=
ion@yahoo.com> wrote:=0A=0AI am considering building a system as follows:=0A=
=0A1. Data stored in Cassandra=0A2. Webservice cluster (stateless) will pu=
ll data from cassandra and do business =0Aoperations plus security enforce=
ment=0A3. Clients will hit the webservice cluster=0A=0AI'm trying to maint=
ain a low read latency and am worried about the number of =0Ahops. Client =
will hit the webservice. The webservice will hit a random node in =0Athe c=
assandra cluster. The cassandra cluster will then route the the =0Aappropr=
iate node and the data will flow all the way back.=0A=0AHow many of these =
hops can I remove? I would bundle the cassandra and =0Awebservice processe=
s onto each box. If I route the webservice to always go to =0Athe local no=
de, I'll remove one hop. Is it possible to optimize this further so =0Atha=
t the client can use the cassandra routing logic to go to the webservice t=
hat =0Aalso houses a cassandra node that contains the data? In this case, =
there would =0Aonly be one hop, and if the data is used frequently, I it w=
ill likely reside in =0Amemory without requiring a separate caching layer.=
 This is an internal =0Awebservice, so I would be ok with a library on the=
 client side to help with the =0Arouting.=0A=0AIs any of this possible? I =
was looking at the cassandra apis and couldn't =0Afigure out a way.=0A=0AT=
hanks for any help!=0A=0A=0A=0A=0A
--Apple-Webmail-42--6c0a9470-6bd9-ca28-e82a-c9ce23d0f40f
Content-Type: multipart/related;
    type="text/html";
    boundary=Apple-Webmail-86--6c0a9470-6bd9-ca28-e82a-c9ce23d0f40f


--Apple-Webmail-86--6c0a9470-6bd9-ca28-e82a-c9ce23d0f40f
Content-Transfer-Encoding: quoted-printable
Content-Type: text/html;
    charset=ISO-8859-1;

<html><body><div>No need to worry.&nbsp;</div><div><br></div><div>I run REST requests =
through Varnish box &gt; nginx / Tornaod / Python box &gt; Cassandra clust=
er and can get requests in and out of the stack in a couple of millisecond=
s. Using some old workstation HW and not paying much attention to tuning.&=
nbsp;</div><div><br></div><div>Build it like a normal system and separate =
out the parts, if / when you have problems then you can look at tuning the=
 cassandra cluster or other parts of the stack. There are normally a numbe=
r of other issues to deal with before network IO.</div><div><br></div><div=
>Hope that helps.&nbsp;</div><div>Aaron</div><div><br><br>On 17 Nov, 2010,=
at 03:31 PM, E S &lt;tr1sklion@yahoo.com&gt; wrote:<br><br></div><div><blo=
ckquote type=3D"cite"><div><div class=3D"_stretch">I am considering buildi=
ng a system as follows:<br>=0A<br>=0A1.  Data stored in Cassandra<br>=0A2.=
  Webservice cluster (stateless) will pull data from cassandra and do busi=
ness <br>=0Aoperations plus security enforcement<br>=0A3.  Clients will hi=
t the webservice cluster<br>=0A<br>=0AI'm trying to maintain a low read la=
tency and am worried about the number of <br>=0Ahops.  Client will hit the=
 webservice.  The webservice will hit a random node in <br>=0Athe cassandr=
a cluster.  The cassandra cluster will then route the the <br>=0Aappropria=
te node and the data will flow all the way back.<br>=0A<br>=0AHow many of =
these hops can I remove?  I would bundle the cassandra and <br>=0Awebservi=
ce processes onto each box.  If I route the webservice to always go to <br=
>=0Athe local node, I'll remove one hop.  Is it possible to optimize this =
further so <br>=0Athat the client can use the cassandra routing logic to g=
o to the webservice that <br>=0Aalso houses a cassandra node that contains=
 the data?  In this case, there would <br>=0Aonly be one hop, and if the d=
ata is used frequently, I it will likely reside in <br>=0Amemory without r=
equiring a separate caching layer.  This is an internal <br>=0Awebservice,=
 so I would be ok with a library on the client side to help with the <br>=0A=
routing.<br>=0A<br>=0AIs any of this possible?  I was looking at the cassa=
ndra apis and couldn't <br>=0Afigure out a way.<br>=0A<br>=0AThanks for an=
y help!<br>=0A<br>=0A<br>=0A<br>=0A      <br>=0A</div></div></blockquote><=
/div></body></html>
--Apple-Webmail-86--6c0a9470-6bd9-ca28-e82a-c9ce23d0f40f--

--Apple-Webmail-42--6c0a9470-6bd9-ca28-e82a-c9ce23d0f40f--