ONE BIG INSTANCE IN LARGE SERVER OR MULTIPLE SMALL INSTANCES ON THAT LARGE SERVER

[[email protected]]

Question regarding Efficient Hardware Usage: we have a physical server which is large Scale as per blueprints ( 20 cores CPU , 600 GB mem , 4x10 GB nic , 4 X 4 16 GB hba etc ) ... Problem is as per recommendations its best to use one instance per server but to utilize this much of resources the database size is going to be around 4 TB which is not good either...
so in this kind of scenario where hardware is too big for one instance is it better to have ONE BIG INSTANCES IN LARGE SERVER OR MULTIPLE SMALL INSTANCES ON THATLARGE SERVER

 

[Trident]

Hi,

With the usual rule, your milage may vary.

If you have mostly lots of static data with low change rate, then the system can be split into multiple smaller ones.

 

[marclant]

It's more important that the server or servers be sized properly for the workload.

database size is going to be around 4 TB which is not good either...

Why is that not good?

ONE BIG INSTANCES IN LARGE SERVER OR MULTIPLE SMALL INSTANCES ON THATLARGE SERVER

One large instance. I would not do multiple instance unless there was a legal requirement to separate the clients. For example, you sell backup as a service, Customer1 backups go to Server1, Customer 2 go to Server2.

Otherwise, you are just adding complexity and extra management.

 

[[email protected]]

It's more important that the server or servers be sized properly for the workload.


Why is that not good?


One large instance. I would not do multiple instance unless there was a legal requirement to separate the clients. For example, you sell backup as a service, Customer1 backups go to Server1, Customer 2 go to Server2.

Otherwise, you are just adding complexity and extra management.

Thanks for the update ,

Why is that not good ?

Well , as the database size grows your going to have performance issues with it right i have seen it mentioned
in lot of forums as well.

 

[marclant]

Well , as the database size grows your going to have performance issues with it right i have seen it mentioned
in lot of forums as well.

Only if you don't have the proper storage. And 4 databases of 1 TB or 1 database of 4 TB will require the same IOPS.

Follow the blueprint recommendations for how to carve up the database and you will be fine. I have worked with customers with 8 TB databases, some have terrible performance because of the way they configured the storage, some have had great results because they did things differently.

 

[[email protected]]

Hi,

With the usual rule, your milage may vary.

If you have mostly lots of static data with low change rate, then the system can be split into multiple smaller ones.

so if the change rate is higher you think its better to have one large instance ?

At this point i dont know if our change rate is high or not , for example one of instances total occupancy size is 562 TB ( before dedupe ) our daily change rate is approx 13.5 TB . if there are 2 instances like that then what would be your suggestion

 

[[email protected]]

Only if you don't have the proper storage. And 4 databases of 1 TB or 1 database of 4 TB will require the same IOPS.

Follow the blueprint recommendations for how to carve up the database and you will be fine. I have worked with customers with 8 TB databases, some have terrible performance because of the way they configured the storage, some have had great results because they did things differently.

Thanks marclant ,

for your inputs.

 

[Trident]

so if the change rate is higher you think its better to have one large instance ?

At this point i dont know if our change rate is high or not , for example one of instances total occupancy size is 562 TB ( before dedupe ) our daily change rate is approx 13.5 TB . if there are 2 instances like that then what would be your suggestion

That big node is probably worth a dedicated SP host/instance, 13.5 TB ingest per day is quite high. in particular if that is after clients has done dedup and compression.

If you are going to use multiple instances/host, then your hardware and os may dictate some limitations. AIX/Power and x86 have different options.


I have mostly AIX systems that runs SP. S922 and S822 with 20 cores/2cpu and 512 Gb RAM
These are split into two partitions, each with dedicated nics and hba's. I try to keep my DB size to about 2 Tb per partition. The bottlenecks are mostly disk systems.

For DB size, I try to keep it below 2 Tb.

I know this is a grey answer, but building SP host depends upon your workloads, your available hardware/os/disks/lan/fabric/tape and so on.