cancel
Showing results for 
Search instead for 
Did you mean: 

1 Filesystem 2 machines

SOLVED
Go to solution
E.A
Occasional Advisor

1 Filesystem 2 machines

Hello everybody,

I'm looking for a solution to write a file to a filesystem while mounting the same filesystem to another machine.

So why would you do that ?

We're supporting a 3 Datacenter solution with a desaster side several miles away.
Well - now the question came up, how long it actually takes until the data written to the local filesystem is replicated through XP continous access to the desaster side.
How long does it actually take untile the data is physically available in the desaster side ?

We thought the easiest way probably is, to sync the machines via NTP and constantly write a file with the current time to the filesystem.
Machine B is reading the file and checking the time difference.

I've activated the VG on machineB as read-only (-a r) and mounted the lvol also read-only.
And as I read in some other threads - the data won't be synced because the machine is holding a chached version of the FS.

So I'm not looking for a supported / stable solution to this problem, we just want to do these tests within a testenvironment to check the XP CA performance.

Any hints on that ?

Thank you very much in advance.
Greetings
EA




Change is the transformation of what is already known
10 REPLIES
Sagar Sirdesai
Trusted Contributor

Re: 1 Filesystem 2 machines

Hi
I did perform CA replication on EVA 5K

1. Select the source EVA i.e. Standby EVA
2. Select the Vdisk
3. Go to Data replication tab
4. Click on add member
5. Name the Data replication group
6. Select destination system which is the only option available.
7. Select the destination disk group
8. Select the failsafe mode as disabled.
9. Use the write mode method as synchronous.
10. Click on create.
11. Check the data replication folder icon and check the state for members for copy state which it should show say NORMAL

Please note when the CA replication is progress initially there will heavy I/O activities

Sagar
Suraj K Sankari
Honored Contributor

Re: 1 Filesystem 2 machines

Hi,

I think you need to go with HP-SG, you need go for cluster environment there your expect ion will be fulfill.

Suraj
Duncan Edmonstone
Honored Contributor

Re: 1 Filesystem 2 machines

EA,

Why not just time a pairsyncwait command on the CA pair? With the right options, pairsyncwait can be configured to wait until a write on a CA PVOL has reached the SVOL. Simply put a standard UNIX timex command in front of this and you know how long writes are taking to reach the remote site.

HTH

Duncan

HTH

Duncan
E.A
Occasional Advisor

Re: 1 Filesystem 2 machines

Hello everybody,

thanks for the reply.

I was thinking about the solution to stop the link and measure the time it takes until it is resynchronised.
But this does not reflect the delay while there is disk I/O.
Good input, but not useful in my particular situation.

I think I'm going to look for a solution to access the RAW device.
I will search the RAW device for the offset, where the file is located and try to read this bytes over and overy again.

Does somebody know a good way to search a PV raw device for a particular "string" ?
I need to find out the offset.





Change is the transformation of what is already known
Eric SAUBIGNAC
Honored Contributor
Solution

Re: 1 Filesystem 2 machines

Bonjour,

Following your idea, here is my silly one : why don't you create a dedicated CA pair for that purpose ? As it will be dedicated, you will not have to worry about any offset (that could change in time) or any dammage you could do. On one side, just write a periodic time stamp (or whatever you need, sequence number for example) at the beginning of the disk and wait for it on the other side.

HTH

Eric
Steven E. Protter
Exalted Contributor

Re: 1 Filesystem 2 machines

Shalom,

I think you should run the test you are proposing in your latest post. It is essential to know the performance of data replication from main site to disaster.

Understand however that the amount of transactions while replication is turned off will effect the test.

If the system is idle during the disconnect test, replication will catch up quickly.

Essential during this test is to insure there is a reasonable to heavy volume of transactions to disk at the main data center or the test results are not useful.

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
Olivier Masse
Honored Contributor

Re: 1 Filesystem 2 machines

Are you using asynchronous or synchronous replication? If it's synchronous there is no issue here since any write needs to be acknowledged from both XPs before the OS considers any data block as written.

With async, you'll never be able to get a good measurement since the time required to send a block to the remote XP will vary depending on the load on your fiber link.

E.A
Occasional Advisor

Re: 1 Filesystem 2 machines

it's an asynchronous link.
but the variaton in the replication speed is exactly what I want to measure.

So - it's quite ok.

I'm currently writing a binary that hopefully will do what I need ;-)

I will post the solution as soon as it's done.

greetings

Change is the transformation of what is already known
Olivier Masse
Honored Contributor

Re: 1 Filesystem 2 machines

Well if that's what you're looking for and you're keen in C, what you can do is a write() with a specific string to a few block(s) on a raw device and print a write timestamp to the scren, while at the same time you're reading it in a loop on the other side of the CA until the string you're looking for appears, then print the read timestamp. You're probably looking for microseconds here so the current date/time won't be enough.

You won't have to bother with the read-only filesystem not being in sync. I did something similar to this once a few years ago, and it worked.

Duncan Edmonstone
Honored Contributor

Re: 1 Filesystem 2 machines

EA,

>> I was thinking about the solution to stop the link and measure the time it takes until it is resynchronised.
>> But this does not reflect the delay while there is disk I/O.
>> Good input, but not useful in my particular situation.

But pairsyncwait *doesn't* stop the link! It simply looks at the current write point on your local XP, and waits until the remote XP reaches that same write point. The upshot of which is it tells you how long it takes for an IO to reach the remote cache.

Here's the relevant section of the RAID Manager XP manual:

The pairsyncwait command is used to confirm that a mandatory write (and
all writes before it) has been stored in the DFW (write) cache area of the
RCU.
The command gets the latest P-VOL async-CA sequence # of the main
control unit (MCU) side file and the sequence # of the most recently
received write at the RCU DFW (with the correct CTGID, group or
raw_device) and compares them at regular intervals.
If the RCU sequence # exceeds the value of the designated MCU sequence
# within the time specified, this command reports a 0 return code (meaning
P-VOL/S-VOL synchronization to the desired point is achieved).

HTH

Duncan

HTH

Duncan