- Community Home
- >
- Servers and Operating Systems
- >
- Integrity Servers
- >
- Re: rx7620 memory allocation
Categories
Company
Local Language
Forums
Discussions
Forums
- Data Protection and Retention
- Entry Storage Systems
- Legacy
- Midrange and Enterprise Storage
- Storage Networking
- HPE Nimble Storage
Discussions
Discussions
Discussions
Forums
Forums
Discussions
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
- BladeSystem Infrastructure and Application Solutions
- Appliance Servers
- Alpha Servers
- BackOffice Products
- Internet Products
- HPE 9000 and HPE e3000 Servers
- Networking
- Netservers
- Secure OS Software for Linux
- Server Management (Insight Manager 7)
- Windows Server 2003
- Operating System - Tru64 Unix
- ProLiant Deployment and Provisioning
- Linux-Based Community / Regional
- Microsoft System Center Integration
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Discussion Boards
Community
Resources
Forums
Blogs
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
тАО09-21-2005 01:39 AM
тАО09-21-2005 01:39 AM
rx7620 memory allocation
I'm shortly going to upgrade both my nodes of rx7620 to 6x CPU from 4 and am wondering about the memory allocation on a per cpu basis.
Currently I have multiple databases running in size of 100GB each and some v.large tables ( in the GB range ). This is an online transaction system where perofrmance is paramount.
When I upgrade to 6cpu, I'll have to re-distribute the memory accross the CPU's ??
so that each CPU will be able to access 4GB. Whereas currently each cpu ( under a 4x system ) can access 6GB RAM.
By the virtue of increasing the number of CPU's in the system and as a side effect reducing the amount of RAM available to each CPU from 6-4GB , combined with the knowleged that I have multiple v.large databases, will the performance of the system be affected in a negative way ..... or is the memory available to all cpu and physical location unimportant ?
If it is cpu specific memory and there is a lot of page swapping going on when using large tables , is the lack of ram per cpu a potential bottleneck and would it be more benefit to install more memory than CPU ?
Any insight into this subject would be appreciated.
Paul.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
тАО09-21-2005 05:56 PM
тАО09-21-2005 05:56 PM
Re: rx7620 memory allocation
something unclear to me here, why are you redistributing the memory when you move from a 4 CPU to a 6 CPU RX7620 ?
When doing so, it means you are probably adding a extra cell board which will have 2 CPU's right ?
Each cell board needs to have memory on it, so are you planning on removing memory DIMMs from the first cell board to install on the second board then during the upgrade?
If the above upgrade scenario is what is taking place at your site, then why not buy the new cell with some memory on it so that you can keep working with 100% CLM (local memory to the CPU's on their own cell board).
HTH
Kris
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
тАО09-21-2005 09:39 PM
тАО09-21-2005 09:39 PM
Re: rx7620 memory allocation
I'm not an expert in Windows memory management on multiprocessor machines, but basically, the memory is shared by all active CPUs in a hardpartition.
As Kris already outlined, it is possible to tune the ammount of (cell-)interleaved memory and cell local (non interleaved) memory to avoid some overhead of cell-cell transfers if independent processes are running on a multicell system which access only small memory ranges. But even if you only use cell local memory, still the complete memory is accessible by each CPU.
But, if you add two CPUs, then I could imagine that you will also increase the number of database licences and processes and if the workload increases and the memory does not increase, then indeed each process would get less of the physical memory than before. But this would also be true if you simply start more processes on the same amount of CPUs.
=> increasing the number of CPUs does not impact the memory performance unless the number of memory using processes increase
best regards
Stefan
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
тАО09-22-2005 05:27 AM
тАО09-22-2005 05:27 AM
Re: rx7620 memory allocation
Any CPU in the rx7620 can access any memory in the rx7620. As such, you do not _need_ to add RAM when you add CPUs.
It is possible you may _want_ to add RAM as you add CPUs. Depends on how much free RAM you had before and whether increasing the throughput of your system necesitates a larger working set.
You may want to look into use and support for Cell Local Memory. By tieing different databased to different Cells and using CLM you can make your N-CPU system look more like N separate systems, which generally is a good thing for scaling. It may also not be required - one of those infamous "it depends" sorts of things. It may even be one of those things were getting some professional consulting might be helpful.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
тАО09-22-2005 07:09 AM
тАО09-22-2005 07:09 AM
Re: rx7620 memory allocation
Everything above is true, IF you are in one big partition. If you have cut Cell 0 into it's own partition, then the cpus on Cell 0 only see the memory on Cell 0 and the same applies to cell 1. If you further carve the partitions down into VPARs, you can only access the amount of memory assigned to the VPAR.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
тАО09-23-2005 12:40 AM
тАО09-23-2005 12:40 AM
Re: rx7620 memory allocation
Anyway, when you go to 2 cell boards it absolutely makes sense to distribute the CPUs and memory evenly across the 2 boards. Introducing the 2nd cell board basically turns the system into a NUMA (non-uniform memory access) box - this means that memory latency is different when accessing 'local' (in the same cell) and 'remote' (in another cell) memory. These NUMA features can lead to some interseting performance issues, so HP introduced the capability within Integrity systems to specify a portion of a cell boards memory as 'cell local memory' - cell local memory is dedicated to just the processors in that cell.
Now for W2K3, HP reccomend that you setup 100% of the memory on a cell as 'cell local memory', I'm not sure why but I'm guessing that W2K3 doesn't perform well in a NUMA configuration. This means that in your configuration although Rick is correct when he says that Itanium memory isn't tied to the CPU, following the HP reccomended guidelines you will effectively have 2 x (3 CPUs with 12GB memory) enforced not by the processor, but by the chipset. If you don't think that's enough I guess you have 2 choices:
1. Ignore the HP reccomendation and set CLM at less than 100% - whether this leads to better performance or not I don't know.
2. Buy some more memory - it's probably cheaper than when you first bought the box, but still not that cheap!
For reference, have a look at this doc:
http://docs.hp.com/en/5990-8230/5990-8230.pdf
The reccomendation for 100% CLM is on page 12.
HTH
Duncan
I am an HPE Employee
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
тАО09-26-2005 09:05 AM
тАО09-26-2005 09:05 AM