[c-nsp] GSR12008|GRP-B|4OC12/ATM-MM-SC|3GE-GBIC-SC throughput?

Lamar Owen lowen at pari.edu
Thu Apr 16 22:45:52 EDT 2009


On Wednesday 15 April 2009 11:34:34 Lamar Owen wrote:
> On Tuesday 14 April 2009 18:22:03 Jason Lixfeld wrote:
> > For the life of us, we can't seem to get any more than 60Mbps
> > sustained across the ATM testing with iperf, so we're just trying to
> > figure out if the GSR just can't push any more than what it's doing or
> > if there's something else afoot.

> I have a 12012 here in production, and have some of the kit necessary to
> test point to point ATM connections (including a Catalyst 8540MSR with
> OC12, ARM, and gigabit cards), and have a 4xOC12/ATM/MM, but it will be a
> few days before I could have the time to set up a test to see if the 12012
> is limited.  

Ok, had a little time today, so got some data. Setup:

Dell Inspiron 600m w/ Gigabit ethernet, running Fedora 10's iperf to a server, 
which is a CentOS 4 VM on an eight-way Opteron VMware ESX system (Dell 
PowerEdge 6950).  

GSR has a 4xOC12 MM ATM card.  

Other ATM OC12 endpoint is a Catalyst 8540CSR with an OC12 ATM MM uplink card 
and a dual GigE card (while I have an 8540MSR, the setup is more complex with 
the MSR than with the CSR with the ATM uplink, and I wanted the simplest 
possible setup to see if the GSR was a limiter).  

As the server is in production, I left it attached to the server farm core 
Extreme Summit1i's, which are GigE-attached to the 12012 GSR.  In the topology 
below, I only list one Summit1i, but there are two in an ESRP setup.

Topology:
600m <-->8540CSR GigabitEthernet10/0/0 via 1000Base-T GBIC
8540CSR ATM0/0/0.1 (VPI/VCI 1/17 PVC) <--> 12012 ATM7/0.1 (VPI/VCI 1/17 PVC)
12012 GigabitEthernet4/0 <--> Extreme Summit1i port 8
Extreme Summit1i port 1 <--> Dell 6950 ESX server GE1.

12012 and the Summit1i are in production (the 12012 is the working side of our 
APS protected OC3 WAN link, and the Summit1i is half of the server farm core), 
and had other traffic, with variable traffic on the VM during test.  I'm pretty 
happy with how much traffic the Dell 600m laptop generated, by the way!

12012 ATM7 is a 4xOC12 ATM MM LC, 8540CSR ATM0/0/0 is a Catalyst 8540 OC12 ATM 
uplink module.  IOS on 12012 is 12.0(32)S12, on the 8540CSR it's 12.1(27b)E3

12012 has two GRP-B's.


Data:
12012 throughput at peak:
pari-gsr-12#sh int atm7/0 load

      Interface                   bits/sec     pack/sec
 --------------------           ------------  ----------
 AT7/0                 Tx          206605000      24617
                       Rx          354535000      34717
pari-gsr-12#

8540CSR throughput at peak:
sr1-8540c-1>sh int atm0/0/0 summ

 *: interface is up
 IHQ: pkts in input hold queue     IQD: pkts dropped from input queue
 OHQ: pkts in output hold queue    OQD: pkts dropped from output queue
 RXBS: rx rate (bits/sec)          RXPS: rx rate (pkts/sec)
 TXBS: tx rate (bits/sec)          TXPS: tx rate (pkts/sec)
 TRTL: throttle count

  Interface              IHQ   IQD  OHQ   OQD  RXBS RXPS  TXBS TXPS TRTL
------------------------------------------------------------------------
* ATM0/0/0                 0     0    0     0 207281000  24491 353708000  
34530
* ATM0/0/0.1               -     -    -     -     -    -     -    -    -
NOTE:No separate counters are maintained for subinterfaces
     Hence Details of subinterface are not shown
sr1-8540c-1>

Output of iperf at client (Dell Inspiron 600m, Pentium M 1.8GHz, Fedora 10), 
slightly sanitized:
[root at localhost ~]# iperf --client esx-host -t 720 --dualtest
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 85.3 KByte (default)
------------------------------------------------------------
------------------------------------------------------------
Client connecting to esx-host, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[  5] local 10.250.132.30 port 46676 connected with esx-host port 5001
[  4] local 10.250.132.30 port 5001 connected with esx-host port 45629
[ ID] Interval       Transfer     Bandwidth
[  4]  0.0-719.9 sec  18.0 GBytes    215 Mbits/sec
[ ID] Interval       Transfer     Bandwidth
[  5]  0.0-720.0 sec  31.3 GBytes    374 Mbits/sec
[root at localhost ~]#


Output of iperf on server (2vCPU VM on a four-way dual core Opteron 2.8GHz 
Dell 6950 ESX 3.5U3; VM running CentOS 4):

[root at esx-host ~]# iperf --server
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 85.3 KByte (default)
------------------------------------------------------------
[  4] local esx-host port 5001 connected with 10.250.132.30 port 46676
------------------------------------------------------------
Client connecting to 10.250.132.30, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[  5] local esx-host port 45629 connected with 10.250.132.30 port 5001
Waiting for server threads to complete. Interrupt again to force quit.
[ ID] Interval       Transfer     Bandwidth
[  5]  0.0-720.0 sec  18.0 GBytes    215 Mbits/sec
[ ID] Interval       Transfer     Bandwidth
[  4]  0.0-720.1 sec  31.3 GBytes    374 Mbits/sec
[root at esx-host ~]#

Port configs:
GSR:
interface ATM7/0                                              
 no ip address                                                
 no ip directed-broadcast                                     
 atm clock INTERNAL                                           
 no atm enable-ilmi-trap
 no atm ilmi-keepalive
!
interface ATM7/0.1 point-to-point
 ip address 10.250.132.25 255.255.255.252
 no ip directed-broadcast
 no atm enable-ilmi-trap
 snmp trap link-status
 pvc 1/17
 !
!

Catalyst 8540CSR:
interface ATM0/0/0
 no ip address
 atm clock INTERNAL
 sonet ais-shut
 arp timeout 900
!
interface ATM0/0/0.1 point-to-point
 ip address 10.250.132.26 255.255.255.252
 pvc 1/17
 !
!

That is pretty good throughput for a single workstation attaching over a GigE 
throttled through an ATM OC12 with AAL5 overhead (SAR, VPI/VCI cell tax, etc) 
to a fairly busy server.

You might find 
http://www.osti.gov/bridge/servlets/purl/764365-05obbP/native/764365.pdf and 
http://www-didc.lbl.gov/Talks/GBN.final.pdf to be interesting reading.

In light of LBNL's experience, detailed in those two papers, I'm very happy 
indeed with the results of the laptop test.

Hope that helps.



More information about the cisco-nsp mailing list