[j-nsp] SRX650 Cluster throughput

[Jeff Rooney]

Single flow using iperf for all of my tests, only variation is that
I'm using both udp and tcp tests, and via the cluster udp seems to
handle better, but still only seeing 500Mbit


On Tue, Apr 24, 2012 at 11:46 AM, Stefan Fouant
<sfouant at shortestpathfirst.net> wrote:
> On 4/24/2012 11:36 AM, Jeff Rooney wrote:
>>
>> So I'm doing some testing with an SRX 650 cluster(11.2R6.3) and am
>> starting to see some odd throughput issues, being that this is my
>> first SRX cluster, I'm most likely overlooking something minor.
>>
>> Physical setup is pretty basic, each srx has multiple links connected
>> into its own switch, as well as two cross connects between the pair of
>> SRX for control and fabric. The switches are connected with a 4 port
>> portchannel. All interfaces are gig and negotiated to 1000m. Switches
>> are catalyst 4948's, if that matters.
>>
>> SRX-node0 ------------sw0
>>     | |                       | | |
>> SRX-node1 ------------sw1
>>
>> I have two servers connected into sw0, when on the same vlan iperf udp
>> tests show 900Mbits/sec and tcp tests show 940+Mbits/sec....so far so
>> good. Moving one box to another vlan and a seperate reth, so traffic
>> now traverses srx-node0...traffic plummets. Iperf udp shows
>> 500Mbits/sec and tcp 317Mbits/sec. Prior to setting up the cluster I
>> tested via a single SRX and saw 900Mbits/sec+ for both tcp and udp.
>> Both hosts are on the same switch and traversing node0 which is also
>> on the same switch. Each vlan terminates on its own reth.
>>
>> Any suggestions as to where to look next?
>
>
> What are you using to generate traffic? Is it a single flow by any chance?
>
> --
> Stefan Fouant
> JNCIE-SEC, JNCIE-SP, JNCIE-ENT, JNCI
> Technical Trainer, Juniper Networks
>
> Follow us on Twitter @JuniperEducate