You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@cloudstack.apache.org by Bryan Tiang <br...@hotmail.com> on 2022/11/25 17:30:58 UTC

Advice on Hardware Rack Design for Cloudstack

Hi Cloudstack Community,

Thanks for the help on the previous question regarding how to deploy instances on 2 different physical servers within a cluster.

I would like some advice from the community regarding our rack design, its our first time setting up a Private Cloud.

We plan to start off our cloud stack journey with a single rack (42u). We have come up with a visual to help understand bit better.



This is what we were thinking:

  *   All Infrastructures are Fully Redundant (Eg. Network, Server, Storage etc)

  *   Racks connected to two different PDU Power Sources ( Max 6KW Each)

  *   Each Server is connected to the SAN Storages within the Rack (will not cross out of the Rack)

  *   San Storages shall Mirror Each Other for full storage redundancy.

  *   Additional harddrives to be added to the San Storages as when needed.

  *   Server Capacity shall be configured in the CloudStack Platform to not provision more than 30% (within its cluster) to allow for Auto VM Restart in a new server, should any server fail.

  *   Servers will not have any Local Drives. All Storages are in the San Storage only.

  *   Minimum setup will include minimum 2 Servers.
Additional servers (with same CPU Models) to be added as requirements grow.


We haven’t decided if we are going to be standardising our servers within the rack to be (2x26pCPU) or (2*64pCPU).

Here’s where we need some advice. In both scenarios of the server spec, we aren’t sure what is the recommended (or safe) San Storage spec to go for. We plan to deploy various services in the VMs, but the most intensive would be MSSQL Databases which are very IO Intensive.

Our guys have recommended a 2x50TB Hybrid SSD/HDD San Storage at 150k IOPS for the entire rack. But we would like second opinions as we aren’t sure the specifications are insufficient (Afraid that San Storage would be the bottleneck one day). Ideally would be some sort of san storage spec which we can grow the IO over time as the number of servers in the rack increase (We are starting with 2, plan to increase to max 14 over time).

Any advice on this area? Even comments on our rack design would be great! We really want to start this on the right footing.

Regards,
Bryan






RE: Advice on Hardware Rack Design for Cloudstack

Posted by Alex Mattioli <Al...@shapeblue.com>.
Hi Bryan,


  *   San Storages shall Mirror Each Other for full storage redundancy.

I’d highly recommend placing the mirror SAN in a separate rack.


  *   Additional harddrives to be added to the San Storages as when needed.
Do you plan for disk shelves in the same rack?


  *   Server Capacity shall be configured in the CloudStack Platform to not provision more than 30% (within its cluster) to allow for Auto VM Restart in a new server, should any server fail.

I’m not sure I get this.
Based on a principle of N+1 redundancy if you have 2 hosts you can use each up to 50% capacity, 3 hosts =  66% capacity, 4 hosts = 75% capacity, etc…etc…  I don’t see any scenario where using only 33% capacity makes any sense.

Where do you plan to run your management servers and DB from? This same rack or outside it?

>We haven’t decided if we are going to be standardising our servers within the rack to be (2x26pCPU) or (2*64pCPU).
I’d recommend something in between. Also depends a lot on how much RAM you’ll add those servers, you want to make sure you have a balanced use of CPU/RAM (keep in mind that you can run ACS with 3:1 CPU overcommitment, but RAM you want to keep 1:1 typically)

>Our guys have recommended a 2x50TB Hybrid SSD/HDD San Storage at 150k IOPS for the entire rack. But we would like second opinions as we aren’t sure the specifications are insufficient (Afraid that San Storage would be the bottleneck one day). Ideally would be some sort of san storage spec which we can >grow the IO over time as the number of servers in the rack increase (We are starting with 2, plan to increase to max 14 over time).

3,000 IOPS per TB sounds quite low nowadays. But that really depends on your end-users usage patterns. But as a finger in the air I’d aim for a peak of 10k IOPS per TB.

Also, if you leave 1KW for switching (LAN and SAN) plus 2kw for the storage arrays, that leaves you with 3kw for your hosts, which means 8-10 hosts tops, will be hard to run 14. (but hard to say without knowing the switches and storage specs)

I’d also highly recommend using 25G networking on your hosts.

Cheers
Alex


From: Bryan Tiang <br...@hotmail.com>
Sent: 25 November 2022 18:31
To: users@cloudstack.apache.org
Subject: Advice on Hardware Rack Design for Cloudstack

Hi Cloudstack Community,

Thanks for the help on the previous question regarding how to deploy instances on 2 different physical servers within a cluster.

I would like some advice from the community regarding our rack design, its our first time setting up a Private Cloud.

We plan to start off our cloud stack journey with a single rack (42u). We have come up with a visual to help understand bit better.



This is what we were thinking:

  *   All Infrastructures are Fully Redundant (Eg. Network, Server, Storage etc)
  *   Racks connected to two different PDU Power Sources ( Max 6KW Each)
  *   Each Server is connected to the SAN Storages within the Rack (will not cross out of the Rack)
  *   San Storages shall Mirror Each Other for full storage redundancy.
  *   Additional harddrives to be added to the San Storages as when needed.
  *   Server Capacity shall be configured in the CloudStack Platform to not provision more than 30% (within its cluster) to allow for Auto VM Restart in a new server, should any server fail.
  *   Servers will not have any Local Drives. All Storages are in the San Storage only.
  *   Minimum setup will include minimum 2 Servers.
Additional servers (with same CPU Models) to be added as requirements grow.


We haven’t decided if we are going to be standardising our servers within the rack to be (2x26pCPU) or (2*64pCPU).

Here’s where we need some advice. In both scenarios of the server spec, we aren’t sure what is the recommended (or safe) San Storage spec to go for. We plan to deploy various services in the VMs, but the most intensive would be MSSQL Databases which are very IO Intensive.

Our guys have recommended a 2x50TB Hybrid SSD/HDD San Storage at 150k IOPS for the entire rack. But we would like second opinions as we aren’t sure the specifications are insufficient (Afraid that San Storage would be the bottleneck one day). Ideally would be some sort of san storage spec which we can grow the IO over time as the number of servers in the rack increase (We are starting with 2, plan to increase to max 14 over time).

Any advice on this area? Even comments on our rack design would be great! We really want to start this on the right footing.

Regards,
Bryan