[Beowulf] GPFS on Linux (x86)

[Geoff Jacobs]

Craig Tierney wrote:
> Mark Hahn wrote:
>>> required that you buy their storage.  Also the licensing of
>>> GPFS was based on the quantity of storage, which I didn't like.
>>
>> HP's version of Lustre (SFS) is also capacity-licensed
>> (and I agree, it's a customer-hostile policy.)
> 
> Yes, I was aware of this as well.  I just blanked on it in
> my first email.
> 
>>
>>> I really don't seem many people discussing the good and bad things
>>> about the current crop of distributed/shared filesystems.  Do
>>> they sign a contract saying they can disclose any information about
>>> their operation?
>>
>> well, if you spend significant money on a commercial product,
>> and are using/depending on it, it's not attractive to embarass
>> the vendor in public.
>>
>>> IBM GPFS
>>> Ibrix
>>> Isilon
>>> Terrascale
>>> Netapp
>>> (Did I forget some).
>>
>> Lustre/SFS and Redhat GFS, certainly.
> 
> Sorry, of course Lustre should have been on this list.
> Does GFS work in distributed mode?  I didn't add it
> to the list because since it is a shared filesystem
> (multiple clients access the block devices directly)
> I didn't include it.  If we include GFS, then we
> also have to include at least SGI CXFS and ADIC StorNext.
> 
> Craig

What sort of HA features does GFS have?

GFS and Lustre are both free software, so base cost is essentially free,
but configuring and maintaining are on your bill. GPFS, etc, have list
cost plus support contract costs plus time.

How does the installation cost (time) of Lustre compare to
installation/tuning cost (time) of NFS and the cost (time+money) of the
other cluster file systems. What about ongoing support costs? What does
the supported version of Lustre from HP offer in terms of installation
expertise and long-term support?

-- 
Geoffrey D. Jacobs

Go to the Chinese Restaurant,
Order the Special