RDMA vs TCP/IP = latency war ...
You must check latency in test with another tool with multi-user configurations.
I'm also test it.
By increase client count, there are huge deffrence in latency factor on same throughput usage.
BR,
Jae-Hoon Choi
RDMA vs TCP/IP = latency war ...
You must check latency in test with another tool with multi-user configurations.
I'm also test it.
By increase client count, there are huge deffrence in latency factor on same throughput usage.
BR,
Jae-Hoon Choi
Hi Grant,
Please see the following user manual Mellanox ConnectX-4/ConnectX-5 NATIVE ESXi Driver for VMware vSphere 6.5.
Please see section 3.1.4 Priority Flow Control (PFC) :
Thanks,
Karen.
Karen, Thank you for the reply.
Can you tell me what is Mellanox's recommended configuration if there is just 1 traffic flow? My connectx-4 cards are direct connected between two servers and used only to connect the iser initator to the iser target. There is no other traffic on these nics. I have seen references to creating a vlan and assigning it to one of the priority queues yet none of this applies to my scenario. Should i just run global pause?
You mean you connect QSFP to a CX4 card ?
No, the 40GBaseSR4 QSFP has an MPO12 fibre connector which is then split into 4 independent streams (LC fibre) coming from different devices which use 10GBaseR. I would like to receive data from all streams (transmit is not important).
No unfortunately that does not work.
The run flash_self_safe forces the switch to boot from secondary and we get the exact same error output.
From uboot I attempted to download an image by tftp but the file transfer begins, the switch outputs an error and boots to the same error.
I opened the unit and there are 4 red LEDs so I suspect a hardware failure.
The LEDS are as follows:
Top row of LEDS (next to RAM module, below the chassis fans)
D104 + D105 on RED
LEDS in bottom right of chassis
CPLD 2 R643 - D87 RED
CPLD 4 R645 - D89 RED
The LEDs turn red shortly after power is applied.
Do you know what may have failed? Will the failed components be replaceable as this is a legacy unit and this with another 7 switches may have a similar problem.
Kind regards.
Rav.
In this case, I think you need RMA the switch if the switch under warranty
Hello my configuration is simple OEL 7.4 two Mellanox ConnectX-3 VPI cards , SX1036 switch and two very Fast NVMe drives.
So my problem is that I configured NFS over RDMA using Infiniband Support packages from OEL because OFED from Mellanox not support
NFS over RDMA from version 3.4 + .
Everything is working I can connect to server over RDMA and I can write/read from NFS server etc. but I have a problem with performance .
I done test on my stipe LV and fit shows me 900k IOPS and around 3,8 GB/s using 4k but when I do the same tests on NFS client I can't get more
then 190k IOPS ? Problem is not the bandwidth because when I change the block size I can get even over 4GB/s but the problem seams to be number
of IOPS delivered from server to client.
I am asking maybe somebody have idea ?? I already change size and size to 1m but without any performance benefits.
My next steps will be configure Aggregation (LCAP) to see if it change something , now I'm using only one Port .
Adam
Legacy equipment out of warranty.
Are we able to purchase a service contract or is this equipment unsupported?
Many thanks.
You should contact sales guys
Thank you for the assistance.
This call can now be closed.
Hi Francois,
Please check https://community.mellanox.com/docs/DOC-2964
and unmerged yet patchset https://review.openstack.org/#/c/504911/
Hello Rav,
Mellanox do not seel servicecontracts on 4036E switches anymore. The product is at EOL stage.
for more information, please refer to our EOL info page at: http://www.mellanox.com/page/eol
Sorry we couldn't assist you.
Thanks
Hi Lenny
I can’t access this document.
Can you please send me the PDF version.
There seems to be something wrong with my Mellanox account.
Regards
Francois Kleynhans
I also cannot access the documents listed...
From the output prints you have presented, it looks like your SX6036 switch should be good & supports 56Gb (FDR)
- cables & nics are also fine and capable of fdr
- cables present: Infiniband speeds : SDR , DDR , QDR , FDR
- switch presents:
Supported LLR speeds : FDR10, FDR - which indicates you can set it to 56Gb
Supported speeds : sdr, ddr, qdr, fdr10 - here you see that fdr is missing
so, in my view - all you have to do is to run the following command on the switch that will add FDR on "supported speeds" list
run on CLI commnd:
(config) # interface ib <interface#> speed sdr ddr qdr fdr10 fdr force
(config) # configure write (to save the changes)
you should now see: Supported speeds : sdr, ddr, qdr, fdr10, fdr
this should enable you to uses 56Gb on the switch and on the nics as well
Please try now.
Thanks I can access it now
Hi,
Adding the parameter didn't help. It still gives the same error:
athena:~ # nvme discover -t rdma -a 192.168.0.1 -s 4420
Failed to write to /dev/nvme-fabrics: Invalid argument
athena:~ # dmesg |tail -1
[ 1408.720843] nvme_fabrics: unknown parameter or missing value 'hostid=a61ecf3f-2925-49a7-9304-cea147f61ae' in ctrl creation request
athena:~ # nvme connect -t rdma --nr-io-queues=32 -a 192.168.0.1 -s 4420 -n mcx
Failed to write to /dev/nvme-fabrics: Invalid argument
athena:~ # !dm
dmesg |tail -1
[ 1437.914081] nvme_fabrics: unknown parameter or missing value 'hostid=a61ecf3f-2925-49a7-9304-cea147f61ae' in ctrl creation request
HPE Part Number 712498-B21 is SX6036T FDR10 switch.
Therefore in this case user can't use FDR14 56Gb speed in fabric.
HPE Support document - HPE Support Center
Why did you give a incorrect information to community users?
Best Regard,
Jae-Hoon Choi