HDS Synchronous True Copy extension question

Unanswered Question
Apr 5th, 2007
User Badges:
  • Bronze, 100 points or more

Greetings,


I was wondering if anyone has used a MDS and ONS 15454 solution to extend the distance limit for HDS sync true copy.


It seems that native fibre has a best practice limit of about 32 km for intensive true copy usage for OLTP. Non intensive supposedly can go up to about 200 km. After that, the latency will cause major problems.


I recall reading somewhere that using DWDM along with the MDS can offer some quite extensive distance expansion to the order of a couple of thousand km's. What I did not see was anything to overcome the latency issue that is there with distance.


I believe the speed of light in fibre is about 5 nanoseconds per metre. So say a 1000 km link would take about 20 milliseconds for the storage cache at both ends to complete the transaction.


So is there any special "thing" available when using the MDS and the ONS to improve the distance by reducing the latency? I know that fibre channel write acceleration is an option but does it overcome the latency issue?


Cheers


Stephen

  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
colin.mcnamara Thu, 04/05/2007 - 07:00
User Badges:
  • Bronze, 100 points or more

I think what you are referring to is the buffer to buffer credit spoofing. This effectively creates a protocol proxy at each mds / ons peering point. The ONS will reply with the R_RDY acknowledgments instead of having this messages pass through the entire wan link.


Here is a doc that has a nice overview -

http://www.cisco.com/en/US/products/hw/optical/ps2011/products_white_paper0900aecd80181093.shtml


The maximum I have ever really seen talked about for synchronous replication has been around 300 kilometers. At 1000km I would generally be thinking about an async replication strategy. Though at the end of the day it is your call for understanding how your applications use your storage.


If you found this helpful please give a rating.

inch Wed, 04/11/2007 - 22:27
User Badges:
  • Bronze, 100 points or more

G'day Guys,


The r_rdy wont really help with latency, it will only help if your device/s have a small buffer credit capacity.


GEN 2 line cards can have upto 4000 buffer credits which equates to (at 2 credits per KM @ 4gb = 2000km) which equates to 8mb of data down the pipe. (My numbers could be a little off so dont crucify me!)


But then, you still have all that latency.

Async is the go for that kind of distance if your application cant handle a lot of latency.


I am just waiting the arrival of some copper SFP's for my IPS line cards. Once these arrive, I will be setting up some FCIP tunnels over a STM1 and doing some experiments. I will share my results once I'm done if people are keen!


Cheers


Andrew

stephen2615 Wed, 04/11/2007 - 23:47
User Badges:
  • Bronze, 100 points or more

Fellow legends,


HDS offer some strange three data centre solution with a mixture of sync and async replication. I would assume that they would do sync within a 100 or so km's and then async to wherever the other major data centre is located. I have not looked into it but I get a feeling that sooner or later, we have to look at another city for the major DR site. Problem with Canberra is that it is 300 kms from any major location.


The one thing that worries me a bit with the MDS/ONS spoofing with buffer credits is what happens if the sending MDS/ONS also dies at the same time as the storage (bomb goes off).


A UPS is not gonna do much then.


It is probably highly likely that the traffic in transit is transaction log and that's the last thing you would want to lose.


How's that lab going?


BTW, Cisco teased me the other day saying that something big is about to be released with the MDS. I said that making IVR part of the base product would do me just fine....


Cheers


Stephen

colin.mcnamara Thu, 04/12/2007 - 07:37
User Badges:
  • Bronze, 100 points or more

Lab practice is going good. I am in RTP the 8th of May... So that is coming soon.


My big worry of course is the pass fail statistics and the lab availability.

If the average engineer passes on the third try, and it takes 6 months to a year to get a seat, it will be a very long time before I get my number ;)


That being said.. I am really hitting it hard in the lab, pretty much embracing the saying " the more you sweat in peace, the less you bleed in war"


As for the pipeline, From what I have seen I am confident that Cisco will be continuing its growth in the storage market for some time.

The next couple years in storage are going to be pretty darn exciting.


--Colin





inch Thu, 04/12/2007 - 15:58
User Badges:
  • Bronze, 100 points or more

I agree with the spoofing...


Get gen 2 line cards then there is no need to spoof anything and it wont matter if a bomb goes off as the data down the pipe has not been ack'd on the original array until the replicated write has occurred :)


SANos licensing sucks... :( I want IVR and the ability to create my own roles without needing the freaking ent license (at around $16k per 9509)

Actions

This Discussion

 

 

Trending Topics: Storage Networking