Performance issues; new to ceph and petasan
Pages: 1 2
DividedByPi
32 Posts
August 7, 2020, 12:57 pmQuote from DividedByPi on August 7, 2020, 12:57 pmJust wanted to update this. I had the environments a bit misconfigured. I am now seeing much better performance than Ceph-ISCSI (tcmu runner) with Petasan. This is very cool!
I am also really impressed with the CIFS integration to the dashboard... Something even mainline Ceph does not yet do!.
If possible, I would definitely like to hear more about multiple VM scaling off of 1 RBD question that I asked in my previous comment!
Just wanted to update this. I had the environments a bit misconfigured. I am now seeing much better performance than Ceph-ISCSI (tcmu runner) with Petasan. This is very cool!
I am also really impressed with the CIFS integration to the dashboard... Something even mainline Ceph does not yet do!.
If possible, I would definitely like to hear more about multiple VM scaling off of 1 RBD question that I asked in my previous comment!
Last edited on August 7, 2020, 12:58 pm by DividedByPi · #11
admin
2,930 Posts
August 7, 2020, 2:31 pmQuote from admin on August 7, 2020, 2:31 pmIt is up to you if you prefer having single rbd image per vm or an rbd images for multiple vms, it depends on how you want to administer them.
In terms of performance: every thread/io operation to the same rbd image / iSCSI disk is processed separately and concurrently. In some cases there could be limits from the client side, from example in VMWare there is a limit on the ESXi host of 32 concurrent io operations per datastore, this is done to prevent a single ESXi host from using all iops of the cluster for itself, however these set values are still good enough and can be changed if you really need.
It is up to you if you prefer having single rbd image per vm or an rbd images for multiple vms, it depends on how you want to administer them.
In terms of performance: every thread/io operation to the same rbd image / iSCSI disk is processed separately and concurrently. In some cases there could be limits from the client side, from example in VMWare there is a limit on the ESXi host of 32 concurrent io operations per datastore, this is done to prevent a single ESXi host from using all iops of the cluster for itself, however these set values are still good enough and can be changed if you really need.
ghbiz
76 Posts
August 12, 2020, 10:55 amQuote from ghbiz on August 12, 2020, 10:55 am
Quote from DividedByPi on August 7, 2020, 12:57 pm
Just wanted to update this. I had the environments a bit misconfigured. I am now seeing much better performance than Ceph-ISCSI (tcmu runner) with Petasan. This is very cool!
I am also really impressed with the CIFS integration to the dashboard... Something even mainline Ceph does not yet do!.
If possible, I would definitely like to hear more about multiple VM scaling off of 1 RBD question that I asked in my previous comment!
DividedByPi,
Can you please share what MIS-CONFIGURATION did you have as we are trying to track down a performance issue on one of our clusters and cant figure it out.
Quote from DividedByPi on August 7, 2020, 12:57 pm
Just wanted to update this. I had the environments a bit misconfigured. I am now seeing much better performance than Ceph-ISCSI (tcmu runner) with Petasan. This is very cool!
I am also really impressed with the CIFS integration to the dashboard... Something even mainline Ceph does not yet do!.
If possible, I would definitely like to hear more about multiple VM scaling off of 1 RBD question that I asked in my previous comment!
DividedByPi,
Can you please share what MIS-CONFIGURATION did you have as we are trying to track down a performance issue on one of our clusters and cant figure it out.
DividedByPi
32 Posts
August 24, 2020, 12:49 pmQuote from DividedByPi on August 24, 2020, 12:49 pmSure, sorry about the late reply.
I don't know how much help its going to be to you though. I didn't have Round Robin configured on VMware with 1 IOP switching. Once I set that my throughput immediately nearly doubled.
Sure, sorry about the late reply.
I don't know how much help its going to be to you though. I didn't have Round Robin configured on VMware with 1 IOP switching. Once I set that my throughput immediately nearly doubled.
Pages: 1 2
Performance issues; new to ceph and petasan
DividedByPi
32 Posts
Quote from DividedByPi on August 7, 2020, 12:57 pmJust wanted to update this. I had the environments a bit misconfigured. I am now seeing much better performance than Ceph-ISCSI (tcmu runner) with Petasan. This is very cool!
I am also really impressed with the CIFS integration to the dashboard... Something even mainline Ceph does not yet do!.
If possible, I would definitely like to hear more about multiple VM scaling off of 1 RBD question that I asked in my previous comment!
Just wanted to update this. I had the environments a bit misconfigured. I am now seeing much better performance than Ceph-ISCSI (tcmu runner) with Petasan. This is very cool!
I am also really impressed with the CIFS integration to the dashboard... Something even mainline Ceph does not yet do!.
If possible, I would definitely like to hear more about multiple VM scaling off of 1 RBD question that I asked in my previous comment!
admin
2,930 Posts
Quote from admin on August 7, 2020, 2:31 pmIt is up to you if you prefer having single rbd image per vm or an rbd images for multiple vms, it depends on how you want to administer them.
In terms of performance: every thread/io operation to the same rbd image / iSCSI disk is processed separately and concurrently. In some cases there could be limits from the client side, from example in VMWare there is a limit on the ESXi host of 32 concurrent io operations per datastore, this is done to prevent a single ESXi host from using all iops of the cluster for itself, however these set values are still good enough and can be changed if you really need.
It is up to you if you prefer having single rbd image per vm or an rbd images for multiple vms, it depends on how you want to administer them.
In terms of performance: every thread/io operation to the same rbd image / iSCSI disk is processed separately and concurrently. In some cases there could be limits from the client side, from example in VMWare there is a limit on the ESXi host of 32 concurrent io operations per datastore, this is done to prevent a single ESXi host from using all iops of the cluster for itself, however these set values are still good enough and can be changed if you really need.
ghbiz
76 Posts
Quote from ghbiz on August 12, 2020, 10:55 amQuote from DividedByPi on August 7, 2020, 12:57 pmJust wanted to update this. I had the environments a bit misconfigured. I am now seeing much better performance than Ceph-ISCSI (tcmu runner) with Petasan. This is very cool!
I am also really impressed with the CIFS integration to the dashboard... Something even mainline Ceph does not yet do!.
If possible, I would definitely like to hear more about multiple VM scaling off of 1 RBD question that I asked in my previous comment!
DividedByPi,
Can you please share what MIS-CONFIGURATION did you have as we are trying to track down a performance issue on one of our clusters and cant figure it out.
Quote from DividedByPi on August 7, 2020, 12:57 pmJust wanted to update this. I had the environments a bit misconfigured. I am now seeing much better performance than Ceph-ISCSI (tcmu runner) with Petasan. This is very cool!
I am also really impressed with the CIFS integration to the dashboard... Something even mainline Ceph does not yet do!.
If possible, I would definitely like to hear more about multiple VM scaling off of 1 RBD question that I asked in my previous comment!
DividedByPi,
Can you please share what MIS-CONFIGURATION did you have as we are trying to track down a performance issue on one of our clusters and cant figure it out.
DividedByPi
32 Posts
Quote from DividedByPi on August 24, 2020, 12:49 pmSure, sorry about the late reply.
I don't know how much help its going to be to you though. I didn't have Round Robin configured on VMware with 1 IOP switching. Once I set that my throughput immediately nearly doubled.
Sure, sorry about the late reply.
I don't know how much help its going to be to you though. I didn't have Round Robin configured on VMware with 1 IOP switching. Once I set that my throughput immediately nearly doubled.