Mailing-List: contact users-help@trafficserver.apache.org; run by ezmlm
Precedence: bulk
Reply-To: users@trafficserver.apache.org
Received-SPF: pass (nike.apache.org: domain of Steve.Owens@disney.com
 designates 204.128.192.17 as permitted sender)
From: "Owens, Steve" <Steve.Owens@disney.com>
To: "users@trafficserver.apache.org" <users@trafficserver.apache.org>
Date: Fri, 13 Apr 2012 13:10:10 -0700
Subject: Request for suggestions failure to scale in cluster mode.
Thread-Topic: Request for suggestions failure to scale in cluster mode.
Thread-Index: Ac0ZsW1vRke1/YmcTGqcAKTa2rfbmw==
Message-ID: <CBADD632.12143%steve.owens@email.disney.com>
Accept-Language: en-US
Content-Language: en-US
user-agent: Microsoft-MacOutlook/14.12.0.110505
acceptlanguage: en-US
Content-Type: multipart/alternative;
	boundary="_000_CBADD63212143steveowensemaildisneycom_"
MIME-Version: 1.0

--_000_CBADD63212143steveowensemaildisneycom_
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: quoted-printable

Hello All,

We are benchmarking our ATS configuration which uses a custom ATS plugin.

At any rate we are observing some interesting behavior.

We are configured to hit about 5 urls in traffic server which return a payl=
oad of about 500 bytes.

These URI's are being cached using cache control headers such that the requ=
ests are almost 100% being served from cache.

At any rate we can't seem to push the cluster past 2800 transactions per se=
cond.

The interesting thing is that when we reduce the back end node count from 2=
 to 1, the 1 node maintains the load at about 60% cpu.
When we run with 2 nodes we see about 30% cpu on each node.

Can anyone offer suggestions on how to get this cluster to scale linearly w=
ith respect to number of nodes?

We know that our load test software is capable of generating in excess of 2=
0K tps, so this seems to be some sort of contention in the cluster itself.

Thoughts?

Best Regards,

Steve Owens


--_000_CBADD63212143steveowensemaildisneycom_
Content-Type: text/html; charset="us-ascii"
Content-Transfer-Encoding: quoted-printable

<html><head></head><body style=3D"word-wrap: break-word; -webkit-nbsp-mode:=
 space; -webkit-line-break: after-white-space; color: rgb(0, 0, 0); font-si=
ze: 14px; font-family: Calibri, sans-serif; "><div>Hello All,</div><div><br=
></div><div>We are benchmarking our ATS configuration which uses a custom A=
TS plugin.</div><div><br></div><div>At any rate we are observing some inter=
esting behavior.</div><div><br></div><div>We are configured to hit about 5 =
urls in traffic server which return a payload of about 500 bytes.</div><div=
><br></div><div>These URI's are being cached using cache control headers su=
ch that the requests are almost 100% being served from cache.</div><div><br=
></div><div>At any rate we can't seem to push the cluster past 2800 transac=
tions per second.</div><div><br></div><div>The interesting thing is that wh=
en we reduce the back end node count from 2 to 1, the 1 node maintains the =
load at about 60% cpu.</div><div>When we run with 2 nodes we see about 30% =
cpu on each node.</div><div><br></div><div>Can anyone offer suggestions on =
how to get this cluster to scale linearly with respect to number of nodes?<=
/div><div><br></div><div>We know that our load test software is capable of =
generating in excess of 20K tps, so this seems to be some sort of contentio=
n in the cluster itself.</div><div><br></div><div>Thoughts?</div><div><br><=
/div><div>Best Regards,</div><div><br></div><div>Steve Owens</div><div><br>=
</div></body></html>

--_000_CBADD63212143steveowensemaildisneycom_--