Forums

Home / Forums

You need to log in to create posts and topics. Login · Register

IOPS benchmark is very low

Pages: 1 2

Hi Team,

We have created 4 nodes PetaSAN cluster .

We have tested on 3 nodes IOPS benchmark.

4K IOPS rados benchmark

pool: rdb

Threads/Client: 256

Duration : 5 min

Cluster IOPS

Write Read
1171 2721

Can you please suggest how we can improve performance . We are looking to implement PetaSAN as a production environment.

Thanks in advance,

Abhishek

how many OSDs do you have ?

are they  spinning hdds ?

after 5 min test, look at chart for disk %utilzation, is it high ?

if you have a spare empty disk (non OSD) , can you run the raw disk performance test from blue console menu ?

We have total 8 OSDs UP & IN.

No, They are not spinning HDDs.. We are using SSD.

After 5 min test disk %utilzation is very low...less then 10%

Please find attached below raw disk performance test from blue console menu

 

The random read disk test is very low.

Our random read disk test is very low means there is issue with our disks or any thing else & please suggest what we can do for best performance.

could be many things: the type of disks itself, controller, firmware...

what is the model of SSD you are using ?

the command we run for random disk iops is

fio --name=xx --filename=/dev/sdXX --iodepth=1 --rw=randread --bs=4k --direct=1 --runtime=10 --time_based --numjobs=8 --group_reporting

replace /dev/sdXX with correct name and numjobs=8 with desired threads

Hi Team,

I used ceph  rbd in my project(create rbd and map in client by mounting it as data partition  ,use like a local disk), will the PetaSAN have this similar feature?

Not sure i understand the question, but PetaSAN includes Ceph, so anything you did before with Ceph you can do, open a shell and just type Ceph commands, create rbd images...etc.  Then from a Linux client, you can map the disk and mount it.

Weather we can implement bond ip for PetaSAN setup ?

Please confirm & how we can implement?

You can configure network including bonds via ui when you first setup the cluster, during deployment of first node "Create New Cluster'.

After you deploy, you can change the configuration on a running cluster manually via /opt/petasan/config/cluster_info.json. The format is very easy to understand but it is easier to deploy a 1 node test vm and play with settings/bongs in ui and take its config file and apply change to running cluster.

 

 

Pages: 1 2