Subscribe

FAS 8060 c-mode throughput issue

[ Edited ]

Hi ,

 

I have a fas 8060 cluster 8.3.2 with FC connection LACP 10 g. By the client side we use Ubuntu 14.4 NFS shares. When I test my connection with fio performance test the iops and disk IO workloads I have low througputs than I had expected. I have attached the outputs from sysstat -x , port utilizations and perfstat during the process to the message. 

 

I have 800-900 mb/s with 1 client and  1100-1200 mb/s with 2 client connections. But this values has decreased dramatically to 500-600 mb/s in a few minutes. I am not waiting 2gb/s

 

throughput but this values is unacceptable. I have already checked the switch,cabling,etc. but I didn't find any solution. I have moved the volume to SAS and SSD but I have the same result.

 

Can someone help me? I dont know was I will exact looking or testing? 

 

thanks!

 

 

 

 

 

Re: FAS 8060 c-mode throughput issue

Hello,

 

I don't see any attachments on your post.

 

Please give us more information:

 

- What is the output of following command during the performance reduction?

 

::> set advanced

 

::> dashboard performance show

 

node shell: "sysstat -x 1", "sysstat -m 1" and "sysconfig -r"

 

::> statistics show-periodic -node -object nfsv3 -instance -counter nfsv3_ops| nfsv3_read_ops| nfsv3_write_ops| nfsv3_throughput| write_avg_latency| read_avg_latency| write_percent| read_percent

 

--- change nfsv3 to nfsv4 if you use nfsv4

 

- Do you have one lif per node ?

 

- How much volumes and aggregates do you use for the test ? (disk type, disk count)

 

- Any flash accelerator in use ?

 

- FIO config files

 

Best regards

Re: FAS 8060 c-mode throughput issue

Hi,

 

Now we have extra workloads on the System like CIFS share etc. I can provide the outputs of commands but I think that will not help to you. 

 

I have attached the logs,when the problem has first time appeared. I have already tried to change to nfs version, but it didn't help. 

Re: FAS 8060 c-mode throughput issue

Hi,

 

The most interesting information will be:

 

- the layout of the aggregates (output: sysconfig -r)

- the count of volumes and aggregates per node which are used for the tests (parallelization is important, because of the many cores...)

- how much transfers from how much server and how long

- output fio config file to know the workload & statistics show-periodic

- is it a synthetic workload or a real life workload ?

 

- "... but this values is unacceptable" What did you expect ?

 

Please also open a case and call your partner to recheck the tests.

 

 

Best regards