[Beowulf] scratch File system for small cluster

2008-09-25 Thread Glen Beane
I am considering adding a small parallel file system ~(5-10TB) my small cluster (~32 2x dual core Opteron nodes) that is used mostly by a handful of regular users. Currently the only storage accessible to all nodes is home directory space which is provided by the Lab's IT department (this is a

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Joe Landman
Glen Beane wrote: I am considering adding a small parallel file system ~(5-10TB) my small cluster (~32 2x dual core Opteron nodes) that is used mostly by a handful of regular users. Currently the only storage accessible to all nodes is home directory space which is provided by the Lab's IT

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Glen Beane
On 9/25/08 10:19 AM, Joe Landman [EMAIL PROTECTED] wrote: Glen Beane wrote: I am considering adding a small parallel file system ~(5-10TB) my small cluster (~32 2x dual core Opteron nodes) that is used mostly by a handful of regular users. Currently the only storage accessible to all nodes

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Joe Landman
Glen Beane wrote: [...] Hi Glen: BLAST uses mmap'ed IO. This has some interesting ... interactions ... with parallel file systems. for what its worth, we use Paracel BLAST and are also considering mpiBLAST-pio to take advantage of a parallel file system Cool.

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Scott Atchley
On Sep 25, 2008, at 10:19 AM, Joe Landman wrote: We have measured NFSoverRDMA speeds (on SDR IB at that) at 460 MB/s, on an RDMA adapter reporting 750 MB/s (in a 4x PCIe slot, so ~860 MB/ s max is what we should expect for this). Faster IB hardware should result in better performance,

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Tim Cutts
On 25 Sep 2008, at 3:19 pm, Joe Landman wrote: BLAST uses mmap'ed IO. This has some interesting ... interactions ... with parallel file systems. It's not *too* bad on Lustre. We use it in production that way. Are there other recommendations for fast scratch space (it doesn't have to

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Joe Landman
Scott Atchley wrote: On Sep 25, 2008, at 10:19 AM, Joe Landman wrote: We have measured NFSoverRDMA speeds (on SDR IB at that) at 460 MB/s, on an RDMA adapter reporting 750 MB/s (in a 4x PCIe slot, so ~860 MB/s max is what we should expect for this). Faster IB hardware should result in

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Huw Lynes
On Thu, 2008-09-25 at 09:40 -0400, Glen Beane wrote: I am considering adding a small parallel file system ~(5-10TB) my small cluster (~32 2x dual core Opteron nodes) that is used mostly by a handful of regular users. Currently the only storage accessible to all nodes is home directory space

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Greg Lindahl
On Thu, Sep 25, 2008 at 10:19:26AM -0400, Joe Landman wrote: BLAST uses mmap'ed IO. This has some interesting ... interactions ... with parallel file systems. The PathScale compilers use mmap on their temporary files. This led to some interesting bugs being reported... fortunately, we were

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Joe Landman
Greg Lindahl wrote: On Thu, Sep 25, 2008 at 10:19:26AM -0400, Joe Landman wrote: BLAST uses mmap'ed IO. This has some interesting ... interactions ... with parallel file systems. The PathScale compilers use mmap on their temporary files. This led to some interesting bugs being reported...

Re: [Beowulf] scratch File system for small cluster

2008-09-25 Thread Greg Lindahl
On Thu, Sep 25, 2008 at 02:08:23PM -0400, Joe Landman wrote: It looks like people use mmap files to explicitly avoid seeks, replacing semantics of file IO with memory access semantics. Well, it explicitly avoids having to call I/O functions all the time as you skip around a file. It also