[bones]

Dear all,

I'am new to this forum so first of all: hello to all.
Hope I can get some information to my questions here.
I have a p570 with:
4 p5+ CPU
8 GB RAM
8 Dualport 4Gbit FC-Adapter
2 Dualport 1Gbit Ethernet Adapter (+2 internal but used for HMC)
RAID:
DS4800 with SATA and FC-Disks in two SAN-Fabrics
Library:
Shared Quantum SCSI-Library


The Problem:
I have to create min. of 5 Partitions to install about 10 TSM-Servers,
two in each partition.
But I don't know what I should do.
Questions:
Use VIO, but howto do virtualization with the FC-Adapter?
We wan't failover for disk I/O and also for Tape I/O.
And different adapters for disk I/O and Tape I/O.
Is something like this possible?
Or do we not need different adapter for disk and tape?
Should we map each physical adapter directly to the partition itself or share via VIO?
Should we gave the disk on the DS4800 to VIO or can we map it directly to the partition itself? And then what about a failover?
Do we get enough Performance with VIO?
What about Failover of VIO, can we or better should we use a second VIO?
Thanks for all replies.

Bones.

[cdelgadop]

Hi

Hmmm. Ok. Let's see if i can help you.

First, your HW config,

You have 4 p5+ CPU which could be micropartitioned if your p5 has the APV feature active (i think so), so you can assign fractions of CPU to each LPAR. In theory you can create up to 40 LPARs with 0.1 CPU each. If you confi VIOS LPAR assign one CPU to it.

You have 8 GB RAM, just remember that some memory will be used by the Hypervisor (Firmware layer controlling all p5 interactions with the LPARs) so you might end with 7GB RAM to assign to those 5 LPARS. If using VIOS LPAR assign at least 512MB and test performance.

You have 8 Dualport 4Gbit FC-Adapter (HBA). Since you want to config LPARs in order to install TSM maybe you want to connect to your Quantum Library using Fibre Channel. VIOS (up to now) cannot virtualizate access to tape devices, so you will need at least 5 HBAs, each for LPAR to tape traffic. You might think in use one port to the Tape SAn and one to the Disk SAN, it's supposed that those ports are independent so is factible to do it, sadly i have no experience with those adapters, please someone comment on this. This lefts only 3 to use for disk traffic (in fact 6 FC connections).

2 Dualport 1Gbit Ethernet Adapter (+2 internal but used for HMC). SO you didn't buy any PCI ethernet adapter ??, remember that p5 servers bring 2 internal ethernet adapters apart from those for HMC connections. Please confirm this.

Let's see your questions.

Use VIO, but howto do virtualization with the FC-Adapter?: You can use those 3 HBA and assign them to the VIOS LPAR. Then design and config the SAN with your 2 fabrics, your DS4800 (with two controllers) and your 3 HBAs so maximum redundancy and performance will be achieved and create the LUNS and hosts configurations neccesary to give adecuate storage to all those TSM servers. Once you have created LUNs you can zone all of then to those 3 HBAs, this way your VIOS server will see many hdisk devices you can then virtualize and assign to each LPAR as Virtual SCSI Server Adapters. (easy to say, hard to do).

We wan't failover for disk I/O and also for Tape I/O. And different adapters for disk I/O and Tape I/O. Is something like this possible? Or do we not need different adapter for disk and tape? : Sorry, Do you want failover or not ??. If you want, failover for Disk I/O is implemented at SAN level when designing the proper cabling between your 2 fabrics and your DS4800. Is not a good practice to mix Tape traffic with disk one, there's a redbook that explains this better than i can do.

Should we map each physical adapter directly to the partition itself or share via VIO? You have to map each HBA used for tape traffic to the client LPAR and think about what to do with the rest. Will your TSM LPARs use disk as storage pools ? if so, then you need to use VIOS.

Should we gave the disk on the DS4800 to VIO or can we map it directly to the partition itself? And then what about a failover?: You can do whatever the 2 options, only concern is that mapping a LUN directly to the client LPAR requieres at least one HBA (two if failover is a need), so you will need 10 HBA just for Disk I/O traffic.

Do we get enough Performance with VIO?: Nice question, i'm having performance problems with my VIOS with the SEA Ethernet device, looks like i need to assign more CPU power. Hmmm, i don't know, maybe is not a very good idea to configure servers of high use as VIOS clients. Please any comment about this.

What about Failover of VIO, can we or better should we use a second VIO?: it's the best practice, but remember that you don't have so many resources.

Hope this helps

[mshulman1980]

Sorry I didn't respond sooner. I wasn't 100% sure on virtualizing tape devices on the latest version of VIO and I didn't want to give you bad information.
If you're going to virtualize disk allocation, you need to be very careful about your scheduling across the TSM servers. If you have two FC adapters devoted to a single VIO for connection to your SAN (redundant) , then you'll risk seriously bottlenecking your operation if you have more then oneTSM server doing a large backup or restore at the same time. I've typically gone the route of one medium to large sized server with all hardware resources for TSM. It's not terribly processor intensive, just storage and interface intensive. If you're virtualizing all storage, which bear in mind will make your life more complicated in it adds steps to the process
Also, I agree with cdelgadop that you'll want to seriously consider going with only 1 VIO server to save on resources given your situation. You probably won't be helped much by a same hardware failover for your VIO.
You should also very seriously consider purchasing more NICs for the VIO server to handle the throughput of 5 TSM servers. Personally I would want at least a 4 way gigabit etherchannel with a failover of a 2 way etherchannel if possible.
You might want to consider consolodating some of your TSM servers. Doing so (like going from 5 to 2) might help you maximize your resource utilization on your frame, It will also help you coordinate scheduling concens a little bit better.
Virtualization servers are great as long as you realize that you're relying on the fact that not everything is going to call the same resource at the same time.

[astralvoid]

One way would be to log in to one of the LPARs and run the 'topas -C' command. This will show you some performance statistics of all of the managed system's LPARs and their micropartition entitlements.