This paper introduces basic clustering components and software tools that can be used to build a Linux cluster on IBM eServer OpenPower 710 by utilizing its 64-bit POWER5 architecture. This paper is intended as an introduction for someone interested in building a new Linux cluster running on OpenPower 710 servers or converting an existing cluster to OpenPower 710 servers.
High-Performance and High-Availability Clustering
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
It is really interesting to see, how IBM tries to enter high-volume server market with their current CPU. The only interesting thing about these servers is their partitioning capibility, which allows up to 10 hardware partitions/processor. So you can consolidate several servers with low load on this single machine, just like zSeries is able to consolidate lot of servers.
Are there any other reasons for using these machines? The software situation is probably comparable with Solaris x86, this time there is a lot of IBM software, but hardly software from other vendors. By the way how is the status of 64-bit Linux on PowerPC (or POWER in this case) compared with 64-bit Linux on x86-64? When I read this forum I still have a feeling that there are still several issues with these versions, mostly driver-related, but case of OpenPower that should be solved by IBM, since this is a well-defined architecture.
Another question is the meaning of OpenPower? Does that mean that other companies can also use POWER5 for their servers? Or does IBM provide detailed specifications of the hardware, so lot of OSes can be adopted for this hardware. It would be a real curiosity to see OpenSolaris booting on OpenPower :-).
Just a few thoughts.
Regards,
Anton
The only interesting thing about these servers is their partitioning capibility, which allows up to 10 hardware partitions/processor. So you can consolidate several servers with low load on this single machine, just like zSeries is able to consolidate lot of servers.
It is the best part and as far as i know what you are talking about is virtual partitions in software which if i am not correct limited to AIX. The hard partitions are per cpu i believe and i thos instances i beleive you can run linux or aix.
1. I doubt four boxes and a management node can be called “High Performance Cluster”.
2. Distributed file systems are applicable to a very limited field of scenarios only. One of them is an environment with multiple _dedicated_ I/O-nodes and a library making use of more than just its striping capabilities, e. g. MPI I/O ROMIO on top of PVFS, the other being an environment with so many nodes that a single nfs-server (even with a quite good raid behind it) may have serious problems with concurrent accesses.
3. Logival Volume Management doesn’t have anything to do with clusters directly; the high availability side of LVM isn’t of much use unless you use specialized tools like an online resizer for your filesystem.
4. In which kind of way does the article tell us anything useful about high availability? Heartbeating? Come on. Some links or explanations about e. g. SNMP or network management tools like nagios, opennms or openview would be nice, or maybe checkpointing capable libraries (openAIS) would have been more useful.
Sorry for flaming, but I found the article to be pretty superficial and, in its own way, too specialized on some example components. Anyone who is in the position to administrate a cluster won’t find anything useful in this article, and for everybody else new to this topic, it was far too specialized (what about an introduction to parallel file systems in general, e. g. PVFS, PFS, Vesta/PIOFS etc., network management tools, cluster interconnects like Infiniband or Myrinet, MPIs capabilities?)
– joe
“It is the best part and as far as i know what you are talking about is virtual partitions in software which if i am not correct limited to AIX. The hard partitions are per cpu i believe and i thos instances i beleive you can run linux or aix.”
Wrong. You can do SMTs, MicroPartitioning, DLPARs, Virtual I/O, and Virtual LAN on OpenPower using Linux. Specifically SLES, and RHEL.
The OpenPower system cannot run AIX as it is to cater to the low price Linux Solution. If you want to run AIX, get the pSeries servers.
Regards.
Another question is the meaning of OpenPower? Does that mean that other companies can also use POWER5 for their servers?
You can also buy POWER boxes from Hitachi and Bull. You can even get Federation from Hitachi. IBM seems to partner with everyone!