Mailing List Archive: XMR 10G speed issue

XMR 10G speed issue

Jan 23, 2014, 6:40 AM

Post #1 of 6 (2721 views)

Hello all,

We replaced one of our Brocade RX devices with a Brocade XMR. Beside
some small syntax changes we copied its config (some basic BGP and OSPF)
to the XMR. We now see the issue that from 6+ Gbit on a 10GE port
single-session transfers (simple wget) slow down below 1GE speeds. The
port can and will reach 8-9 Gbit without packet loss, but single-session
wgets will slow down further. We currently have to keep them below
6.5Gbit to prevent any possible impact.

Note: this is in a LAG, our non-LAG ports are currently not pushing
enough traffic to test this quickly.

Anyone an idea what we are missing / forgetting?

We tried:

- Downgrade / Upgrade software (from 5.5d to 5.4e and 5.6a)
- Shutdown 1 switch fabric (tried both options)
- Switchover to standby management
- Single port new lag on unused 10G module

Total amount of traffic has no impact, when removing 2 uplinks from 30G
LAG early in the morning we also see this issue.

Thanks in advance.

Regards,
Rens Ariens
Yisp

Re: XMR 10G speed issue [ In reply to ]

rens at yisp

Jan 23, 2014, 7:21 AM

Post #2 of 6 (2672 views)

Permalink

Just to add: cpu and lp-cpu usage are < 5%.

Regards,
Rens Ariens
Yisp

On 01/23/2014 03:40 PM, Rens Ariens - YISP wrote:
> Hello all,
>
> We replaced one of our Brocade RX devices with a Brocade XMR. Beside
> some small syntax changes we copied its config (some basic BGP and
> OSPF) to the XMR. We now see the issue that from 6+ Gbit on a 10GE
> port single-session transfers (simple wget) slow down below 1GE
> speeds. The port can and will reach 8-9 Gbit without packet loss, but
> single-session wgets will slow down further. We currently have to keep
> them below 6.5Gbit to prevent any possible impact.
>
> Note: this is in a LAG, our non-LAG ports are currently not pushing
> enough traffic to test this quickly.
>
> Anyone an idea what we are missing / forgetting?
>
> We tried:
>
> - Downgrade / Upgrade software (from 5.5d to 5.4e and 5.6a)
> - Shutdown 1 switch fabric (tried both options)
> - Switchover to standby management
> - Single port new lag on unused 10G module
>
> Total amount of traffic has no impact, when removing 2 uplinks from
> 30G LAG early in the morning we also see this issue.
>
> Thanks in advance.
>
> Regards,
> Rens Ariens
> Yisp
>
>
> _______________________________________________
> foundry-nsp mailing list
> foundry-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/foundry-nsp

Re: XMR 10G speed issue [ In reply to ]

rens at yisp

Feb 10, 2014, 4:58 AM

Post #3 of 6 (2630 views)

Permalink

Nobody an idea? Only hardware part we did not replace/exclude yet is the
chassis, but can a defect at the backplane cause such issue?

On a non-LAG port we see the same issue btw.

Regards,
Rens Ariens
Yisp

On 01/23/2014 03:40 PM, Rens Ariens - YISP wrote:
> Hello all,
>
> We replaced one of our Brocade RX devices with a Brocade XMR. Beside
> some small syntax changes we copied its config (some basic BGP and
> OSPF) to the XMR. We now see the issue that from 6+ Gbit on a 10GE
> port single-session transfers (simple wget) slow down below 1GE
> speeds. The port can and will reach 8-9 Gbit without packet loss, but
> single-session wgets will slow down further. We currently have to keep
> them below 6.5Gbit to prevent any possible impact.
>
> Note: this is in a LAG, our non-LAG ports are currently not pushing
> enough traffic to test this quickly.
>
> Anyone an idea what we are missing / forgetting?
>
> We tried:
>
> - Downgrade / Upgrade software (from 5.5d to 5.4e and 5.6a)
> - Shutdown 1 switch fabric (tried both options)
> - Switchover to standby management
> - Single port new lag on unused 10G module
>
> Total amount of traffic has no impact, when removing 2 uplinks from
> 30G LAG early in the morning we also see this issue.
>
> Thanks in advance.
>
> Regards,
> Rens Ariens
> Yisp
>
>
> _______________________________________________
> foundry-nsp mailing list
> foundry-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/foundry-nsp

Re: XMR 10G speed issue [ In reply to ]

niels=foundry-nsp at bakker

Feb 10, 2014, 5:24 PM

Post #4 of 6 (2635 views)

Permalink

* rens@yisp.nl (Rens Ariens - YISP) [Thu 23 Jan 2014, 16:08 CET]:
>We replaced one of our Brocade RX devices with a Brocade XMR. Beside
>some small syntax changes we copied its config (some basic BGP and
>OSPF) to the XMR. We now see the issue that from 6+ Gbit on a 10GE
>port single-session transfers (simple wget) slow down below 1GE
>speeds. The port can and will reach 8-9 Gbit without packet loss,
>but single-session wgets will slow down further. We currently have
>to keep them below 6.5Gbit to prevent any possible impact.
>
>Note: this is in a LAG, our non-LAG ports are currently not pushing
>enough traffic to test this quickly.
>
>Anyone an idea what we are missing / forgetting?

How many switch fabric modules do you have? The symptoms sound as if
you have one too few. You should see that in "show sfm-utilization
all". And check that you have enough lanes up with "show sfm-links
all | i DOWN" although you hint that you rebooted a few times so that
should be ok.

-- Niels.

--
_______________________________________________
foundry-nsp mailing list
foundry-nsp@puck.nether.net
http://puck.nether.net/mailman/listinfo/foundry-nsp

Re: XMR 10G speed issue [ In reply to ]

mkallen at gmail

Feb 10, 2014, 5:30 PM

Post #5 of 6 (2637 views)

Permalink

Niels is thinking the same thing I was, it seems like a SFM problem, or
possibly a traffic manager on a module. If you have a support contract, I
would say open a ticket. I can't think of anything in the config that
would be that different from RX to MLX, other than maybe jumbo frames?

Mike

On Mon, Feb 10, 2014 at 5:24 PM, Niels Bakker
<niels=foundry-nsp@bakker.net>wrote:

> * rens@yisp.nl (Rens Ariens - YISP) [Thu 23 Jan 2014, 16:08 CET]:
>
> We replaced one of our Brocade RX devices with a Brocade XMR. Beside some
>> small syntax changes we copied its config (some basic BGP and OSPF) to the
>> XMR. We now see the issue that from 6+ Gbit on a 10GE port single-session
>> transfers (simple wget) slow down below 1GE speeds. The port can and will
>> reach 8-9 Gbit without packet loss, but single-session wgets will slow down
>> further. We currently have to keep them below 6.5Gbit to prevent any
>> possible impact.
>>
>> Note: this is in a LAG, our non-LAG ports are currently not pushing
>> enough traffic to test this quickly.
>>
>> Anyone an idea what we are missing / forgetting?
>>
>
> How many switch fabric modules do you have? The symptoms sound as if you
> have one too few. You should see that in "show sfm-utilization all". And
> check that you have enough lanes up with "show sfm-links all | i DOWN"
> although you hint that you rebooted a few times so that should be ok.
>
>
> -- Niels.
>
> --
> _______________________________________________
> foundry-nsp mailing list
> foundry-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/foundry-nsp
>

Re: XMR 10G speed issue [ In reply to ]

rens at yisp

Feb 10, 2014, 5:47 PM

Post #6 of 6 (2639 views)

Permalink

Niels,

We started with 2/3 SFMs, but we quickly upgraded this to 3/3 and have
tested with all 2/3 combinations. All links show up.

We tested with ports on different 10G modules as well, so traffic
manager on a module should be excluded as well i think.

Thanks for the suggestions, if you got any other idea's please let me know.

Rens

Niels Bakker schreef op 2/11/2014 2:24 AM:
> * rens@yisp.nl (Rens Ariens - YISP) [Thu 23 Jan 2014, 16:08 CET]:
>> We replaced one of our Brocade RX devices with a Brocade XMR. Beside
>> some small syntax changes we copied its config (some basic BGP and
>> OSPF) to the XMR. We now see the issue that from 6+ Gbit on a 10GE
>> port single-session transfers (simple wget) slow down below 1GE
>> speeds. The port can and will reach 8-9 Gbit without packet loss, but
>> single-session wgets will slow down further. We currently have to
>> keep them below 6.5Gbit to prevent any possible impact.
>>
>> Note: this is in a LAG, our non-LAG ports are currently not pushing
>> enough traffic to test this quickly.
>>
>> Anyone an idea what we are missing / forgetting?
>
> How many switch fabric modules do you have? The symptoms sound as if
> you have one too few. You should see that in "show sfm-utilization
> all". And check that you have enough lanes up with "show sfm-links
> all | i DOWN" although you hint that you rebooted a few times so that
> should be ok.
>
>
> -- Niels.
>

_______________________________________________
foundry-nsp mailing list
foundry-nsp@puck.nether.net
http://puck.nether.net/mailman/listinfo/foundry-nsp