[Pvfs2-users] Can PVFS over multiple disks on a single node tolerate disk failures?

[Sam Lang]

On Jan 18, 2008, at 4:44 PM, Scully wrote:

> That's really a wonderful answer. Thank Sam.
>
> Let me ask another question. Since the system I just mentioned will  
> not tolerate disk failure, I will try "Installing PVFS2 on a  
> cluster". Say, I have several nodes, I use one disk for PVFS2 data  
> on each.
>
> Dose such a system tolerate node or server failure? Is there anyway  
> to configure it to achieve this, if one node fails, the application  
> still can get files from other nodes? If it can, how many nodes I  
> have to have at least?

This setup is the same as the RAID0 case again.  PVFS simply stripes  
the data over the nodes, without redundancy, so if a disk fails, you  
can't recover the data.  If you want fault tolerance with PVFS, you  
can use a redundant RAID (RAID1, RAID5, etc.) scheme on *each* node  
over a few disks.

You can provide shared storage (which is of course more expensive) to  
all of the nodes, and then be able to have a PVFS setup that is both  
fault-tolerant and HA capable.

The choice of how to setup PVFS really depends on the hardware (both  
storage and network) that you have (or plan to buy), and the workload  
you have (how you plan to use the filesystem).  We might be able to  
give you more tips on a good setup if we knew what that hardware and  
workload setup looked like.

-sam

>
>
> Thank you.
>
>
>
> On Fri, 2008-01-18 at 15:59 -0600, Sam Lang wrote:
>>
>> On Jan 18, 2008, at 2:29 PM, Scully wrote:
>>
>> > I want to store PVFS data on four disks in the same node. If I  
>> don't
>> > choose RAID, based on my understanding to <
>> 3.2 How can I store PVFS
>> > data on multiple disks on a single node?> in the FAQ, I should set
>> > four servers, each for a disk.
>>
>> >
>> > Can such a system tolerate disk failures?
>>
>> It would be similar to RAID0 over those disks.  There's no mirroring
>>
>> or data redundancy, so a disk failure would cause a loss of data.
>>
>> > Should I set each disk a metadate server to get the disk failure
>> > tolerance?
>>
>>
>> In your case adding metadata servers won't help, since all the  
>> servers
>> are running on the same node.  Just pick one of the disks (servers)  
>> to
>>
>> manage the metadata.
>>
>> >
>> >
>> > I think the <7.3 Can PVFS tolerate disk failures?> in the FAQ  
>> means,
>>
>> > if I use RAID5 or RAID6 to combine the disks first, then, it can
>> > tolerate, am I right?
>>
>> Yes.  redundant RAID schemes provide for recovery if a disk should
>>
>> fail.  Running PVFS servers on-top of RAID5 usually slows down writes
>> a little, but if you want to tolerate failures, the performance hit  
>> is
>>
>> probably worth it.
>> -sam
>>
>> >
>> >
>> > Thank you.
>> >
>>
>> > _______________________________________________
>> > Pvfs2-users mailing list
>> >
>> Pvfs2-users at beowulf-underground.org
>> > http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
>>
>>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.beowulf-underground.org/pipermail/pvfs2-users/attachments/20080118/ae7826b5/attachment.htm