headers
Adrian Saul | 1 Feb 12:27

Re: ZFS and Oracle RAC


No - ZFS is not multi node aware yet.  I don't know much about it but 
there is a product called "lustre" which I think was in early stages of 
supporting ZFS as a global filesystem, so you might want to look at that.

Perhaps you could look at sharing the pool via NFS or as iSCSI volumes 
as well.

Phillip Bruce wrote:
>  
> 
> Hi,
> 
> I am working in a environment that involves Oracle RAC. We are
> implementing 2 clusters with 4 node each.
> The plan is to do an RMAN to ZFS filesystem labeled /backup . The
> /backup filesystem is 12 TB which
> should be sufficient between the 2 clusters.
> 
> I believe in practice that your supposed to deport and import zpools
> between hosts. The plan is to mount /backup ZFS
> filesystem only on node 1 in cluster 1 and node 5 in cluster 2. Is there
> any way I can keep backup mounted on both
> nodes simultaneously without having to deport and import all the time. I
> think the answer will be no because zfs is not
> a global filesystem.
> 
> But I've asked the DBA's if the ORACLE SID will be different on both
> nodes. They said yes and that I like to setup
> the structure so backups to node 1 is done in one directory structure
(Continue reading)

Permalink | Reply | 1 comment |
headers
Adrian Saul | 1 Feb 12:29

Re: CIFS and NFS


That is correct - by default Samba runs in async mode.

Bob Friesenhahn wrote:
>  
> 
> On Sun, 31 Jan 2010, Dick Hoogendijk wrote:
> 
>  > Googling around I get the impression that NFS is much faster in reading
>  > files from the (NFS) server than CIFS from a CIFS server. For writing
>  > this is mostly the other way around; CIFS beats NFS.
>  > Is this in essence true? Because if it is, I should choose NFS for
>  > mostly read-only shares and CIFS for read/write ones. Correct me if I'm
>  > wrong.
> 
> I do see incredibly fast write speed from my Windows system over CIFS,
> even when using Samba. However, the faster CIFS writes surely
> sacrifice reliability and assured consistency of NFS. If the CIFS
> server spontaneously reboots, I would be willing to bet that the files
> are in an uncertain state from the view of the client, which has to
> log in again. Sent data may be lost.
> 
> Bob
> --
> Bob Friesenhahn
> bfriesen <at> simple.dallas.tx.us <mailto:bfriesen%40simple.dallas.tx.us>, 
> http://www.simplesystems.org/users/bfriesen/ 
> <http://www.simplesystems.org/users/bfriesen/>
> GraphicsMagick Maintainer, http://www.GraphicsMagick.org/ 
> <http://www.GraphicsMagick.org/>
(Continue reading)

Permalink | Reply | 1 comment |
headers
Ryan Anderson | 1 Feb 15:27
Picon

Re: ZFS and Oracle RAC

Ditto to what Adian said.

Also, you didn't mention the interconnects involved, which could bring up
different answers. Since this is backups, I'd highly recommend you just use
NFS from the ZFS filesystem, its the most practical and simple. If we were
talking about the type of storage for database data, it would be a different
discussion; but even then I'd highly recommend NFS, but paired with LACP
link aggregation (using dladm).

On Mon, Feb 1, 2010 at 5:27 AM, Adrian Saul <asaul <at> sgtbundy.ods.org> wrote:

>
> No - ZFS is not multi node aware yet.  I don't know much about it but
> there is a product called "lustre" which I think was in early stages of
> supporting ZFS as a global filesystem, so you might want to look at that.
>
> Perhaps you could look at sharing the pool via NFS or as iSCSI volumes
> as well.
>
>
> Phillip Bruce wrote:
> >
> >
> > Hi,
> >
> > I am working in a environment that involves Oracle RAC. We are
> > implementing 2 clusters with 4 node each.
> > The plan is to do an RMAN to ZFS filesystem labeled /backup . The
> > /backup filesystem is 12 TB which
> > should be sufficient between the 2 clusters.
(Continue reading)

Permalink | Reply | 0 comments |
headers
John Martinez | 1 Feb 18:08
Picon

Re: ZFS and Oracle RAC

Phillip,

You are better off using OCFS or bite the bullet and do Veritas SFRAC (CFS+VCS) for this functionality.

-john

On Jan 31, 2010, at 3:18 PM, Phillip Bruce wrote:

> Hi,
> 
> I am working in a environment that involves Oracle RAC. We are 
> implementing 2 clusters with 4 node each.
> The plan is to  do an RMAN  to  ZFS  filesystem  labeled  /backup . The  
> /backup  filesystem is 12 TB  which
> should be sufficient between the 2 clusters. 
> 
> I believe in practice that your supposed to deport and import zpools 
> between hosts. The plan is to mount /backup ZFS
> filesystem only on node 1 in cluster 1 and node 5 in cluster 2. Is there 
> any way I can keep backup mounted on both
> nodes simultaneously without having to deport and import all the time. I 
> think the answer will be no because zfs is not
> a global filesystem.
> 
> But I've asked the DBA's if the ORACLE SID will be different on both 
> nodes. They said yes and that I like to setup
> the structure so backups to node 1 is done in one directory structure 
> and node 5 in another. So in effect you have the
> following:
>
(Continue reading)

Permalink | Reply | 1 comment |
headers
Phillip Bruce | 2 Feb 11:02
Picon

Re: ZFS and Oracle RAC

John,

I wonder now if fstyp command will even recognized OCFS?

Just to answer those that already responded, i thank those that have.

All luns are FC SAN infact, HDS storage arrays USPV more specifically is 
being used.
I have heard of OCFS and Lustre. But have not used either. Currently we 
are also using
Oracle ASM which is neat so far and it uses RAW disks.

I'll have to read up on OCFS. So at this point I'll have to use the 
following sequence:

1. deport /backup from node 5
2. import /backup to node 1
3 when RMAN is complete the deport /backup on node 1
4. then import to node 5 then have RMAN recover data to node5.

You see node 5 is one of the standby nodes in the 2nd cluster.
the idea is to restore data from one of the node in the first cluster.

The 2nd cluster from my understanding is a standby cluster.
It is also a development environment as well.

That the reason for my asking. As for the interconnects, it all GigE. 
Each node has 2 nic cards and using using IPMP.

I'll agree that NFS makes a good way of sharing storage across nodes in
(Continue reading)

Permalink | Reply | 0 comments |
headers
Phillip Bruce | 2 Feb 11:04
Picon

Re: CIFS and NFS

Should reboot happen and your client is only capable of running SMB 1.0 
then your stuck anyway.
Only windows 2008 is capable of SMB 2.0 that solves the issues of client 
disconnect.

phillip

Adrian Saul wrote:
> That is correct - by default Samba runs in async mode.
>
>
> Bob Friesenhahn wrote:
>   
>>  
>>
>> On Sun, 31 Jan 2010, Dick Hoogendijk wrote:
>>
>>  > Googling around I get the impression that NFS is much faster in reading
>>  > files from the (NFS) server than CIFS from a CIFS server. For writing
>>  > this is mostly the other way around; CIFS beats NFS.
>>  > Is this in essence true? Because if it is, I should choose NFS for
>>  > mostly read-only shares and CIFS for read/write ones. Correct me if I'm
>>  > wrong.
>>
>> I do see incredibly fast write speed from my Windows system over CIFS,
>> even when using Samba. However, the faster CIFS writes surely
>> sacrifice reliability and assured consistency of NFS. If the CIFS
>> server spontaneously reboots, I would be willing to bet that the files
>> are in an uncertain state from the view of the client, which has to
>> log in again. Sent data may be lost.
(Continue reading)

Permalink | Reply | 0 comments |
headers
Peter Evans | 2 Feb 14:18
Picon

Re: Dell Poweredge T100, again

On Fri, Jan 29, 2010 at 07:11:38PM -0500, John Taylor wrote:
> +1 for the Intel DP35DP.  Thing has been very solid.  was a little
> weird with ATA/ICHI getting
> it booted the first time, cause I didn't have a SATA DVD drive and
> ended up jumpstarting it.

	I'll throw in my 2yen for pretty much any intel board. 

	Here is a 965 board, you can find 965 in industrial gear. It's not
	blazing but lots of sata, hexagon has six drives.

hexagon 3> prtdiag
System Configuration:
BIOS Configuration: Intel Corp. CO96510J.86A.6043.2008.0224.0740 02/24/2008

==== Processor Sockets ====================================

Version                          Location Tag
-------------------------------- --------------------------
Intel(R) Core(TM)2 CPU          6320  @ 1.86GHz LGA 775

==== Memory Device Sockets ================================

Type    Status Set Device Locator      Bank Locator
------- ------ --- ------------------- --------------------
DDR2    in use 0   J1MY                CHAN A DIMM 0
DDR2    in use 0   J2MY                CHAN A DIMM 1
DDR2    in use 0   J3MY                CHAN B DIMM 0
DDR2    in use 0   J4MY                CHAN B DIMM 1
(Continue reading)

Permalink | Reply | 3 comments |
headers
Mike Riley | 2 Feb 19:34
Picon

Re: Dell Poweredge T100, again

Peter Evans wrote:
>  
> 
> On Fri, Jan 29, 2010 at 07:11:38PM -0500, John Taylor wrote:
>  > +1 for the Intel DP35DP. Thing has been very solid. was a little
>  > weird with ATA/ICHI getting
>  > it booted the first time, cause I didn't have a SATA DVD drive and
>  > ended up jumpstarting it.
> 
> I'll throw in my 2yen for pretty much any intel board.

My issue with the Intel boards is that the memory costs a lot more than 
equivalent speed AMD CPUs require.  Also, the bus contention issues they 
have had compared to what AMD does with it's Hyperchannel configs, 
although I think they were converting to s similar type of bus design. 
Not sure if the latest Intel CPUs have switched to that or not, as I 
haven't followed it closely since I left Sun.

Mike

------------------------------------

Please check the Links page before posting:
    http://groups.yahoo.com/group/solarisx86/links
Post message: solarisx86 <at> yahoogroups.com
UNSUBSCRIBE:  solarisx86-unsubscribe <at> yahoogroups.com
Permalink | Reply | 2 comments |
headers
John D Groenveld | 2 Feb 20:39
Picon

Re: Dell Poweredge T100, again

In message <4B68702E.6030307 <at> cox.net>, Mike Riley writes:
>Not sure if the latest Intel CPUs have switched to that or not, as I 
>haven't followed it closely since I left Sun.

AMD MDEs were going out on sales calls with the sales critters and SEs
hawking John Fowler and company's volume systems.

Were their counterparts at Intel providing the same technical
briefings to Sun customers?

Most importantly, are AMD and Intel in bed with Larry Ellison
and company to hawk volume Solaris systems?

John
groenveld <at> acm.org

------------------------------------

Please check the Links page before posting:
    http://groups.yahoo.com/group/solarisx86/links
Post message: solarisx86 <at> yahoogroups.com
UNSUBSCRIBE:  solarisx86-unsubscribe <at> yahoogroups.com
Permalink | Reply | 0 comments |
headers
Matt Banks | 2 Feb 20:42
Picon

Re: Dell Poweredge T100, again


On Feb 2, 2010, at 11:34 AM, Mike Riley wrote:

> Peter Evans wrote:
> > 
> > I'll throw in my 2yen for pretty much any intel board.
> 
> My issue with the Intel boards is that the memory costs a lot more than 
> equivalent speed AMD CPUs require. Also, the bus contention issues they 
> have had compared to what AMD does with it's Hyperchannel configs, 
> although I think they were converting to s similar type of bus design. 
> Not sure if the latest Intel CPUs have switched to that or not, as I 
> haven't followed it closely since I left Sun.

If you go off of Sun's MSRP (because I know we ALL buy ONLY Sun RAM :) ), you see that DDR3 and DDR2 (for x86, not
SPARC) are at pretty similar price point for server RAM.

1x2GB DDR3-1333 for x4170: $135
1x4GB DDR3-1333 for x4170: $255* 
1x8GB DDR3-1066 for x4170: $1050

2x2GB DDR2-667 for x4140: $220
2x4GB DDR2-667 for x4140: $400
2x8GB DDR2-667 for x4140: $2500

A quick search on crucial showed this:

1x2GB DDR3-1333 for x4170: $95
1x4GB DDR3-1333 for x4170: $184
1x8GB DDR3-1066 for x4170: $749
(Continue reading)

Permalink | Reply | 0 comments |

Gmane