A computer components & hardware forum. HardwareBanter

If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below.

Go Back   Home » HardwareBanter forum » General Hardware & Peripherals » Storage & Hardrives
Site Map Home Register Authors List Search Today's Posts Mark Forums Read Web Partners

SAN question



 
 
Thread Tools Display Modes
  #1  
Old October 20th 03, 10:53 AM
nospam
external usenet poster
 
Posts: n/a
Default SAN question

Hi,
In a SAN enviornment, can two servers have access to the same storage device
at the the same time?
I want to implement a failover, one server doing all the writing, the other
only read from the disk. If one server goes down, the other will do both.
Is this possible and normal way to use SAN?

Where do you configure this, on server, storage or switch?
Thanks


  #2  
Old October 20th 03, 12:09 PM
Henry Newman
external usenet poster
 
Posts: n/a
Default

On Mon, 20 Oct 2003 02:53:31 -0700, "nospam"
wrote:

This is only possible with a shared file system from someone like
ADIC, IBM, SGI, Sun, Veritas and others and sever failover from IBM,
Sun, SGI, Veritas and others. It requires a great deal of work and
knowledge of both the file system, HBA failover, server failover,
networks, switch zoning and RAID issues just to name a few.

If this is a critical system get professional help.

Regards,

Henry



Hi,
In a SAN enviornment, can two servers have access to the same storage device
at the the same time?
I want to implement a failover, one server doing all the writing, the other
only read from the disk. If one server goes down, the other will do both.
Is this possible and normal way to use SAN?

Where do you configure this, on server, storage or switch?
Thanks


  #3  
Old October 20th 03, 06:27 PM
Faeandar
external usenet poster
 
Posts: n/a
Default

Actually, what the poster is asking about does not require a shared
file system, merely server clustering and shared LUN access.

As an example, you can use Veritas Cluster Server (VCS) and Veritas
Volume Manager to manage what you're asking. What I do is mirror with
VM and setup VCS to import the mirror as rw in case of failover.

Given your question though you would not even need to mirror since you
can access the same LUN if the host mounts it ro. It would look
something like this:

Host A - mounting LUN1 as rw. primary host for LUN1
Host B - mounting LUN1 as ro, failover host for LUN1

If Host A barfs Host B re-mounts LUN1 as rw. Where the scripting
takes place really depends on how you're failing over. Is it custom
scripts or commercial package. Either way I guess something triggers
Host B to re-mount LUN1.

~F

On Mon, 20 Oct 2003 02:53:31 -0700, "nospam"
wrote:

Hi,
In a SAN enviornment, can two servers have access to the same storage device
at the the same time?
I want to implement a failover, one server doing all the writing, the other
only read from the disk. If one server goes down, the other will do both.
Is this possible and normal way to use SAN?

Where do you configure this, on server, storage or switch?
Thanks


  #4  
Old October 20th 03, 06:48 PM
Nicolas CHEVE
external usenet poster
 
Posts: n/a
Default

Le Mon, 20 Oct 2003 02:53:31 -0700
"nospam" a =E9crit:

Hi,
In a SAN enviornment, can two servers have access to the same storage
device at the the same time?
I want to implement a failover, one server doing all the writing, the
other only read from the disk. If one server goes down, the other will
do both. Is this possible and normal way to use SAN?

Where do you configure this, on server, storage or switch?
Thanks


Hi,

I'm only working on HP-UX but it would exist the same on several OS.
With HP-UX, you could use MC-ServiceGuard.
This is a cluster application.
A cluster application is the best way to protect the data against
hasardous actions with the RO or RW mount option used manually or with a
"manual cluster mode" using script.
The bad point is that it's not chip....


In conjontion with the LVM cluster mode, you could do the following:

Host A as active nde an Host B as standby node. Their are node of the
cluster, MC Serviguard is installed on both.
Both A and B server could acces to the lun thru the SAN (with zone or
lunmasking).

When the application is active on server A, MC ServiGuard will active
the VolumGroup in LVM with the Cluster Mode.
In this way, on the host B, MC Service Guard will see that the
VolumGroup (and the LUN) are in use on an other node of the cluster (you
could have more than two node) an wil not attempt to acces it. If an
other host(host C) could acces too to the LUN, LVM (even if MC
ServiceGuard is not installed on host C) could see that the VG is in use
in cluster mode and don't allow to acces it in rw mode on host C (maybe
in ro but i'm no sure).

If the host A will crash, then the hearbit (network) will indicate to
host B that it must take the work on.
Host B will become the active node.

@++
Nicolas
  #5  
Old October 20th 03, 08:34 PM
Rene Köhnen-Wiesemes
external usenet poster
 
Posts: n/a
Default

Hi,
In a SAN enviornment, can two servers have access to the same storage

device
at the the same time?
I want to implement a failover, one server doing all the writing, the

other
only read from the disk. If one server goes down, the other will do both.
Is this possible and normal way to use SAN?

Where do you configure this, on server, storage or switch?
Thanks

Most configurations I've seen that are working like this are cluster
configurations,
like other posters have said here. This kind of application running on the
server
itself, and cluster nodes are talking to each other via ip-heartbeat. There
are lots
of different concepts out there, depending on whatever solution you prefer -
you can choose between native OS solutions (e.g HACMP/AIX, MCS/Windows etc.)
In addition there are also 3rd party products available, mostly OS
independed like
Veritas Cluster Manager. For the storage systems you only have to make sure
that
each cluster node can access each LUN.

René


  #6  
Old October 21st 03, 02:48 PM
Bill Todd
external usenet poster
 
Posts: n/a
Default


"Faeandar" wrote in message
...
Actually, what the poster is asking about does not require a shared
file system, merely server clustering and shared LUN access.


I'm afraid not. The OP stated that the second system was not solely being
used for failover, but also would be reading the data concurrently while the
first system wrote it. Unless they're using a shared file system, the
second server will be caching stale data from the disk without knowing that
it has been changed, and will quickly get very confused.

- bill



  #7  
Old October 21st 03, 06:36 PM
Faeandar
external usenet poster
 
Posts: n/a
Default

My take on the read host was that it only needed to see the disk, not
actually serve read requests. If my take is wrong then you are
correct. If not then the setup I mention will work.

As a side note, isn't it a tuneable paramter? Caching I mean. An OS
could be made to not cache read request data. Of course that would
severly impact it's actual funtion but that's not the question...

~F

On Tue, 21 Oct 2003 09:48:11 -0400, "Bill Todd"
wrote:


"Faeandar" wrote in message
.. .
Actually, what the poster is asking about does not require a shared
file system, merely server clustering and shared LUN access.


I'm afraid not. The OP stated that the second system was not solely being
used for failover, but also would be reading the data concurrently while the
first system wrote it. Unless they're using a shared file system, the
second server will be caching stale data from the disk without knowing that
it has been changed, and will quickly get very confused.

- bill



  #8  
Old October 21st 03, 06:55 PM
Bill Todd
external usenet poster
 
Posts: n/a
Default


"Faeandar" wrote in message
...

....

As a side note, isn't it a tuneable paramter? Caching I mean. An OS
could be made to not cache read request data. Of course that would
severly impact it's actual funtion but that's not the question...


While many file systems allow applications to request unbuffered access to
disk, I know of none that allow you to specify that the file system should
itself cache no *metadata* about the disk or the files on it - and that's
where the worst confusion would occur.

(Not that using unbuffered access to metadata would wholly solve the problem
either: the data location or size could change between the time you read
the metadata and the time you acted based upon it. You really do need
inter-node synchronization of changes, even if only one node is performing
them and the other is only reading.)

- bill


  #9  
Old October 21st 03, 09:29 PM
Nik Simpson
external usenet poster
 
Posts: n/a
Default

Bill Todd wrote:
"Faeandar" wrote in message
...

...

As a side note, isn't it a tuneable paramter? Caching I mean. An OS
could be made to not cache read request data. Of course that would
severly impact it's actual funtion but that's not the question...


While many file systems allow applications to request unbuffered
access to disk, I know of none that allow you to specify that the
file system should itself cache no *metadata* about the disk or the
files on it - and that's where the worst confusion would occur.

(Not that using unbuffered access to metadata would wholly solve the
problem either: the data location or size could change between the
time you read the metadata and the time you acted based upon it. You
really do need inter-node synchronization of changes, even if only
one node is performing them and the other is only reading.)

- bill


In addition, even assuming that the server doing writes can do so completely
unbuffered, there is still a good possiblity that the server with read
access will get bad data. Consider the case where a file is created and
written to, there are many things going on:

1. updates to directory structures
2. updates to the file inode
3. actual data written to the file
4. updates to the filesystem freeblock list
5. ...

Logically these operations are "atomic" i.e. from the point of view of an
application they all take place simultaneously, but physically they are a
sequence of operations with varying amounts of parralel execution, on single
system with read/write access this is not a problem, we have filesystems to
handle everything and prevent race conditions. As soon as you introduce
another uncoordinated reader the whole thing goes to hell in a handbasket
because the second reader might try to read from the disk while an operation
is partially complete.

So, if you want safe (i.e. workable access to a shared disk), you must have
some sort of filesystem that prevents race conditions from screwing up what
the reader and writer see.


--
Nik Simpson


  #10  
Old October 22nd 03, 09:36 PM
Mark Cecil
external usenet poster
 
Posts: n/a
Default

When using FibreChannel (looped, or in a SAN fabric, or both), the only real
thing known by the target device (DDM, LUN, vdisk, whatever) is what the
current request is, and what device acted as the initiator for said request.
The device itself does not arbitrate what things can and cannot be done by
which initiator: If a valid operation is requested by a valid initiator,
the unit will perform the operation, NQA, and return the result to the
initiator.

So, it's really the domain of the involved operating systems, the HBA's, and
SAN fabric to enforce policies of common target usage. And,
misconfiguration of any of these has the potential to be disastrous. Henry
mentions that you should enlist pro help for just this reason.

Consider the following oversimplified example:

- Andy has several luns (call them X, Y, Z) allocated via SAN to UN*X
box "foo", and her super-critical-gotta-be-there-25-hours-a-day Oracle
database files are stored on a filesystem there.

- Barney, the junior storage admin, sets up a mapping that also assigns
lun X to "bar", a Windows 2K3 box, without first checking for conflicts.

- The admin for "bar" (just for fun, we'll call him Goober) now sees
that he has a shiny new (read: not recognized as NTFS-partitioned) lun is
available for use, and attempts to format the device.

- Nearly instantaneously, valid LIVE volume information from "foo" is
being overwritten with formatting data from "bar", and there is nothing
illegal about it, from a storage system point-of-view.

- Both machines throw errors, and Andie, who has just lost all of the
current payroll data, reaches for the largest wet noodle in her desk drawer,
with which she will pummel Goober, then Barney, into the cubicle carpeting.

So, multi-initating any device is easy, but really requires an in-depth
knowledge of the involved initiators and the data transport mechanism to
ensure data integrity.

Moral: Haphazard configuration of SAN resources can lead only to violence.


Best regards,

Mark

P.S. Hiya Henry



--
-= Mark Justin Cecil == New Orleans, LA == =-
-=
http://noml.dyndns.org/mark.html =-
-= UNIX/Storage Architecture, Implementation, and Administration =-

"The truth is the truth, no matter what you *believe*"


"Henry Newman" wrote in message
...
On Mon, 20 Oct 2003 02:53:31 -0700, "nospam"
wrote:

This is only possible with a shared file system from someone like
ADIC, IBM, SGI, Sun, Veritas and others and sever failover from IBM,
Sun, SGI, Veritas and others. It requires a great deal of work and
knowledge of both the file system, HBA failover, server failover,
networks, switch zoning and RAID issues just to name a few.

If this is a critical system get professional help.

Regards,

Henry



 




Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump

Similar Threads
Thread Thread Starter Forum Replies Last Post
couple of Dimension XPS Gen4 question Matt Dell Computers 3 March 4th 05 03:20 AM
Pentium question Robert Sligh Overclocking AMD Processors 4 April 27th 04 01:13 PM
Memtest86 on Laptop Question (OT) Ken Fox Overclocking 7 February 5th 04 10:47 PM
Overvolt question... Gambit Overclocking AMD Processors 2 January 26th 04 09:02 PM
Question about LCD and CRT monitors (Samsung variety) Aaron Chan General 3 October 7th 03 02:04 PM


All times are GMT +1. The time now is 05:07 PM.


Powered by vBulletin® Version 3.6.4
Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.
Copyright ©2004-2024 HardwareBanter.
The comments are property of their posters.