Disk Enclosures
1748180 Members
4279 Online
108759 Solutions
New Discussion юеВ

How to estimate the replication bandwith in async replication?

 
yjf6660
New Member

How to estimate the replication bandwith in async replication?

Hi Expert,

I'm planing a DR porject now and confused at bandwidth estimation. Following is our device list.

Prod site: XP12000 with 56GB cache
DR site: XP12000 with 28GB cache

the two site's distance is about 200KM and we have confirmed to use async replication method.

Referred some doc, I know we must get IOPS(WRITE)/Second at Prod XP12000 to estimate bandwidth of replication link. But it is not easy to capture IOPS value, btw,we use B-seires san switch, we could not know where and how to get the IOPS information at SAN switch or storage level. So we have a simple method to estimate bandwith like below:

We use async replication, the RTO and RPO requirement all are 5 minutes. so the maximum bandwidth is (PROD XP12000'S CACHE/300), that is: 56*1024/300=192MB/s=1920Mb/s. IF we use a 2Gb/s link to support replication, then the delay will under 5minutes. is it right? could you give me some advice on DR bandwidth estimation? and how to capture the IOPS data?

So many thanks for your great help.


Richard
2 REPLIES 2
yjf6660
New Member

Re: How to estimate the replication bandwith in async replication?

Hi All,

Is there anyone familiar with this topic? thanks.
Phillip Thayer
Esteemed Contributor

Re: How to estimate the replication bandwith in async replication?

Richard,

HP has a Replication Workload Profiler worksheet that is used when a CA solution is qualified. I am not sure about the XP but in the EVA world we use the EVAPerf to monitor the amount of data going to each LUN and output it to a CSV file. We can then use Excel to get average sizes for each LUN (and consequently each application) plug this information into the Replication Workload Profiler worksheet and play with different bandwidths to get the desired RTO/RPO times.

Get with your local HP partner or HP themselves and ask about doing this.

Phil
Once it's in production it's all bugs after that.