Geeks,
What will be the best way to have Disaster Recovery of site ( File & Print and Messaging Soluton. ). at remote location with approx. 300 km. distance.
Regards, Animesh.
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
Morning Animesh,
--- Animesh Singh an2rhyme@yahoo.com wrote:
Geeks,
What will be the best way to have Disaster Recovery of site ( File & Print and Messaging Soluton. ). at remote location with approx. 300 km. distance.
[snip]
There could be various ways to approach this particular problem.
Is there any constraint on the amount of Bandwidth???. You would have looked at rsync already....any special reason you found it not suitable to the situation??.
Trevor
Regards, Animesh.
===== ( >- -< ) /~\ __ Scaling FLOSS in the __ /~\ | ) / Enterprise : trevor.w@pvision.biz \ (/ | |_|_ \ Call Now: 9820349221 / _|_| ____________________________________/
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
Animesh Singh wrote:
What will be the best way to have Disaster Recovery of site ( File & Print and Messaging Soluton. ). at remote location with approx. 300 km. distance.
One thing you should consider is whether you really want DR for File & Print and Messaging servers. These functions are easily reproducible with proper backup/restore procedures. DR (both hot and cold DR sites) is an expensive proposition. Keep it only for critical business processes e.g. your ERP & CRM systems or your Software Development systems.
Nilesh.
Afternoon Nilesh,
--- Nilesh Chaudhari listbox@nilesh.net wrote:
Animesh Singh wrote:
What will be the best way to have Disaster
Recovery of
site ( File & Print and Messaging Soluton. ). at remote location with approx. 300 km. distance.
One thing you should consider is whether you really want DR for File & Print and Messaging servers. These functions are easily reproducible with proper backup/restore procedures. DR (both hot and cold DR sites) is an expensive proposition. Keep it only for critical business
<snip> Cost chart for same refers.
Site Cost HW equipment Telecommunications Setup time Location Cold site Low None None Long Fixed Warm site Medium Partial Partial/full Medium Fixed Hot site Medium/high Full Full Short Fixed Mobile site High Dependent Dependent Dependent Not fixed Mirrored site High Full Full None Fixed
processes e.g. your ERP & CRM systems or your Software Development systems.
<snip>
Well u mentioned good point. The alternatives, dependent upon the evaluation of the computer function, may include: hot, warm, or cold sites reciprocal agreements, two data centers, multiple computers, service centers, consortium arrangement, vendor supplied equipment, or a combinations of all of these.
My objective is to achive best using Opensource technology.
Nilesh.
Thanks, Reagrds, Animesh.
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
--- Animesh Singh an2rhyme@yahoo.com wrote:
Afternoon Nilesh, Site Cost HW equipment Telecommunications Setup time Location Cold site Low None None Long Fixed Warm site Medium Partial Partial/full Medium Fixed Hot site Medium/high Full Full Short Fixed Mobile site High Dependent Dependent Dependent Not fixed Mirrored site High Full Full None Fixed
processes e.g. your ERP & CRM systems or your Software Development systems.
forget the technology...get the requirements cleared form the functional guys....are they ok if 1 days data is lost? Do they want their apps to be up and running with 2 days or 2 mintues of a disaster? get answers to these questions and the technology will choose itself
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
Evening Bhagwat,
--- kishor bhagwat aaaaarrrgghhh@yahoo.com wrote:
--- Animesh Singh an2rhyme@yahoo.com wrote:
Afternoon Nilesh, Site Cost HW equipment Telecommunications Setup
time
Location Cold site Low None None Long Fixed Warm site Medium Partial Partial/full Medium Fixed Hot site Medium/high Full Full Short Fixed Mobile site High Dependent Dependent Dependent Not fixed Mirrored site High Full Full None Fixed
processes e.g. your ERP & CRM systems or your Software Development systems.
forget the technology...get the requirements cleared form
<snip>
What will be the best way to have Disaster Recovery of
site ( File & Print and Messaging Soluton. ). at remote location with approx. 300 km. distance.
& with 2Mbps link.
the functional guys....are they ok if 1 days data is lost?
<snip> There should be little to no data loss in soultion.
Do they want their apps to be up and running with 2 days or 2 mintues of a disaster?
<snip> It should be like Mirrored Sites, if not then Hot site.
get answers to these questions and the technology will choose itself
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
Reagrds, Animesh.
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
Animesh Singh wrote:
There should be little to no data loss in soultion.
...
What will be the best way to have Disaster Recovery of site ( File & Print and Messaging Soluton. ). at remote location with approx. 300 km. distance. & with 2Mbps link.
There can never be `no data loss'. Some data is always lost in such incidents where a whole site goes down. As Kishor said, you need to decide how much days/hours worth of data you can afford to lose. Given that you need to have a DR site for File & Print servers, Print servers do not store any data permanently. So you can rule them out. File servers will typically have data that is not modified at every instant per se. The data is static over a couple of hours. But the size and amount of data transaction is not known. So one cannot tell whether you need a fibre channel to your DR site or a 64kbps leased line will suffice. Same argument holds for messaging servers.
The alternatives, dependent upon the evaluation of the computer function, may include: hot, warm, or cold sites reciprocal agreements, two data centers, multiple computers, service centers, consortium arrangement, vendor supplied equipment, or a combinations of all of these. My objective is to achive best using Opensource technology.
None of these can be achieved using software technologies. Now comes the question of your main question - open source DR tools. You have these options for Linux/*nix -
1. journalling file system 2. tar, gunzip 3. dd, cp 4. fsck 5. taper, cdrw-taper 6. linux single 7. http://linux-ha.org/ 8. http://www.linuxvirtualserver.org/ 9. and so on...
For Windows you have -
1. System Restore 2. Last known Good configuration 3. Device Driver Roll Back 4. Automated System Recovery 5. Tons of proprietory tools 6. etcetra.
Is that what you wanted? Or are you looking for a one-stop-shop for all your DR needs? IMHO, it is not possible. You have to look at hundreds of parameters before you actually decide how you are going to 1) Plan on a DR site 2) Maintain a DR site 3) Making it work when the d-day arrives. No technology can help you do all that.
Nilesh.
--- Nilesh Chaudhari listbox@nilesh.net wrote:
Animesh Singh wrote:
There should be little to no data loss in soultion.
zero data loss = 'synchronous' solution. anything other than this is async - where ur data loss is unpredictable - can be zero to any number of I/Os.
...
What will be the best way to have Disaster Recovery of site ( File & Print and Messaging Soluton.
). at
remote location with approx. 300 km. distance. & with 2Mbps link.
one 2 Mbps IP link can transfer 2 x 0.8 x 1000 / 8 = 200 Kilobytes of data per second.(0.8 becoz 20% of IP overheads). That means your data at primary should have a change rate much less than this if you want synchronous replication.
There can never be `no data loss'. Some data is always lost in such incidents where a whole site goes down.
Not true. You can have zero data loss solutions - problem is...they need humungous bandwidth and/or they screw up your application response times.
You have these options for Linux/*nix -
- journalling file system
- tar, gunzip
- dd, cp
- fsck
- taper, cdrw-taper
- linux single
- http://linux-ha.org/
- http://www.linuxvirtualserver.org/
- and so on...
none of these will work. you need to use proprietary solutions that will intercept Write calls to the OS, transfer them to DR site, actually write to remote disk, get an acknowledge from there, write to local disk, and *only then* acknowledge a successful write to applcation.(now u know why zero data loss solutions cost a bomb and screw up app response times)
kishor
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
kishor bhagwat wrote:
Not true. You can have zero data loss solutions - problem is...they need humungous bandwidth and/or they screw up your application response times.
Agreed. But practically, do these solutions ensure zero data loss? That would be utopia!
You have these options for Linux/*nix -
- journalling file system
- tar, gunzip
- dd, cp
- fsck
- taper, cdrw-taper
- linux single
- http://linux-ha.org/
- http://www.linuxvirtualserver.org/
- and so on...
none of these will work. you need to use proprietary solutions that will ...
True, proprietory software and hardware based solutions do exist. But that's not what we are looking for. The above options (recovery & availability) are the closest that open source can get to, especially Heartbeat and Failsafe http://oss.sgi.com/projects/failsafe/. Right? Does anyone know about any other open-source solution? I don't know how linux-ha will work over a WAN connection. Never done that. But it should be possible. Google brings up a few interesting links.
Nilesh.
Mornign Kishor,
--- kishor bhagwat aaaaarrrgghhh@yahoo.com wrote:
--- Nilesh Chaudhari listbox@nilesh.net wrote:
Animesh Singh wrote:
There should be little to no data loss in
soultion. zero data loss = 'synchronous' solution. anything other than this is async - where ur data loss is unpredictable - can be zero to any number of I/Os.
...
What will be the best way to have Disaster Recovery of site ( File & Print and Messaging
Soluton.
). at
remote location with approx. 300 km. distance. & with 2Mbps link.
one 2 Mbps IP link can transfer 2 x 0.8 x 1000 / 8 = 200 Kilobytes of data per second.(0.8 becoz 20% of IP overheads). That means your data at primary should have a change rate much less than this if you want synchronous replication.
<snip> But is it fesible to have Site1 synchronous replication with Site2, Site2 synchronous replication with Site3 and Site3 ynchronous replication with Site1 ?
There can never be `no data loss'. Some data is
always
lost in such incidents where a whole site goes down.
Not true. You can have zero data loss solutions - problem is...they need humungous bandwidth and/or they screw up your application response times.
<snip> Very true, Veritas GCM, does same, with IP Bandwidth. This is no specific distance limitation that serves as the delimiter between synchronous & asynchronous modes of operation as it is really a dependency of network architecture ( i.e. SAN, switches, that can support the distance & subjective tolerance to the introduced latency.) A general rule of thumb that is sometimes deployed in enterprises is a latency time of less then 500 milliseconds, round trip.
You have these options for Linux/*nix -
- journalling file system
- tar, gunzip
- dd, cp
- fsck
- taper, cdrw-taper
- linux single
- http://linux-ha.org/
- http://www.linuxvirtualserver.org/
- and so on...
none of these will work. you need to use proprietary solutions that will intercept Write calls to the OS, transfer them to DR site, actually write to remote disk, get an acknowledge from there, write to local disk, and *only then* acknowledge a successful write to applcation.(now u know why zero data loss solutions cost a bomb and screw up app response times)
<snip> What will be the impact of BIC TCP over 2Mbps link. They claim, "BIC can reportedly achieve speeds roughly 6,000 times that of DSL and 150,000 times that of current modems."
kishor
Regards, Aniemsh.
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
--- Animesh Singh an2rhyme@yahoo.com wrote: <snip>
But is it fesible to have Site1 synchronous replication with Site2, Site2 synchronous replication with Site3 and Site3 ynchronous replication with Site1
you can definitely have any such configuration- but can u afford the b/w? Will you have people at 3 diff locations skilled enough to carry recovery processes? Its not practical. Typically you will have site 1 to 2 to 3, with 1 sync-replicating to site 2 which async-replicates to site 3.This is called a multi-hop config and is in use (even in India).
<snip> Very true, Veritas GCM, does same, with IP Bandwidth. This is no specific distance limitation that serves as the delimiter between synchronous & asynchronous modes of operation as it is really a dependency of network architecture ( i.e. SAN, switches, that can support the distance & subjective tolerance to the introduced latency.) A general rule of thumb that is sometimes deployed in enterprises is a latency time of less then 500 milliseconds, round trip.
GCM is a cluster solution(DNS-based). Its not a replication solution. I'm talking only abt data availability at DR site, not app availability (yet). There is no distance limitation for async, in the sense that it will work, but your risk of data loss increases as u go farther. Sync is a diff beast -I wouldnt go over 70 miles - beyond that you're going to hurt your response times.
<snip> What will be the impact of BIC TCP over 2Mbps link. They claim, "BIC can reportedly achieve speeds roughly 6,000 times that of DSL and 150,000 times that of current modems."
I dunno abt BIC. Once the SCSI data packets are in IP, your IP network takes over and you are free to do what u like.
regds, Kishor
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
On 29/04/04 21:30 -0700, kishor bhagwat wrote: <snip>
none of these will work. you need to use proprietary solutions that will intercept Write calls to the OS, transfer them to DR site, actually write to remote disk, get an acknowledge from there, write to local disk, and *only then* acknowledge a successful write to applcation.(now u know why zero data loss solutions cost a bomb and screw up app response times)
Two phase commit. You can always implement a kernel module to do this.
Devdas Bhagat
--- Devdas Bhagat devdas@dvb.homelinux.org wrote:
On 29/04/04 21:30 -0700, kishor bhagwat wrote:
<snip> > none of these will work. you need to use proprietary > solutions that will intercept Write calls to the OS, > transfer them to DR site, actually write to remote disk, > get an acknowledge from there, write to local disk, and > *only then* acknowledge a successful write to > applcation.(now u know why zero data loss solutions cost a > bomb and screw up app response times) Two phase commit. You can always implement a kernel module to do this.
u can. the problem is in handling delays, link failures, and other assorted problems.thats what makes the s/w complex.definitely doable (being done already)..
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
On 29/04/04 00:30 -0700, Animesh Singh wrote: <snip>
Well u mentioned good point. The alternatives, dependent upon the evaluation of the computer function, may include: hot, warm, or cold
Why does this suddenly sound like a homework question to me? Hers a hint: look for DR articles post 9th September 2001. Lots of those around.
Devdas Bhagat
Evening Devdas,
--- Devdas Bhagat devdas@dvb.homelinux.org wrote:
On 29/04/04 00:30 -0700, Animesh Singh wrote:
<snip> > Well u mentioned good point. > The alternatives, dependent upon the evaluation of the > computer function, may include: hot, warm, or cold Why does this suddenly sound like a homework question to me? Hers a
<snip> Its not a homework question for u or for anyone. I am reallying looking into it. and well when U, mentioned about cold & hot sites, it was really great point, as many others could have gain lot from that in this maillist. I have done quite a good study on it and then am trying to put up the points to get more views from geeks on this list. and its really great that am getting very positive feed backs from everyone.
hint: look for DR articles post 9th September 2001.
<snip>
I have gone through those too, before posting on the LUG and well I tried studying from other commercial vendors too. and also I am in touch with developers of who are involved with DR implementation. ( one such mail I have pasted on lug to. )
My hole point is to get more information on DR and to implement and draft a good case study on opensource and this can be the better place where lots of geeks are adding there comments.
Lots of those around.
Devdas Bhagat
__________________________________ Do you Yahoo!? Win a $20,000 Career Makeover at Yahoo! HotJobs http://hotjobs.sweepstakes.yahoo.com/careermakeover
-
Animesh Singh -
Devdas Bhagat -
kishor bhagwat -
Nilesh Chaudhari -
Trevor Warren