mirrorview bandwidth estimate

Mirrorview/Async under the covers uses a combination of Snapview snapshot and SANCopy/Incremental technology.  Snapview tracks the changes between updates and SANCopy/I transfers the updates to the target.  To get a very accurate MV/A bandwidth requirement you'd have to know a couple things.

1.) What is your RPO (Recovery Point Objective). ie: how much data are you willing to lose on the target if there is an outage.

2.) Average write MB/s for the LUNs that you plan to mirror

3.) Data locality of the writes, ie: how many of the writes are overwriting the same block repeatedly within a short time frame

If you have an RPO of say 30 minutes, than you need MV/A to complete an update every 15 minutes.  So you need to know how much data is written to the LUN during the 15 minute interval, and then you need enough bandwidth to transfer that data across within 15 minutes.

#2 is pretty easy to get from the NAR, you just pull up the LUNs, select write bandwidth and then add them together.  You may want to graph them as an area (or stacked) graph together to show the total bandwidth for all the LUNs together.

#3 on the other hand is very difficult to measure.  You are best off assuming there is NO locality, that all writes are to new blocks and MV/A will need to transfer every write that occurs.  In practice this isn't true but it's a safe way to calculate and will give you some headroom.

If you have access to the array and can take a snapshot of each LUN, then you can see how many blocks are tracked during the snapshot session.  The blocks are 64KB so you can figure out how much data would need to be transferred for the amount of time the snapshot was active.  This will also account for the locality.  Just keep in mind that locality only matters for the period of time equal to the update interval.  If the interval is 15 minutes and the same block is overwritten every 30 minutes, you will need to transfer that block every time.  It's only if the block is updated multiple times in the same 15 minute interval that you get bandwidth savings.

Example:   If you take a snapshot and let is run for 15 minutes and you see 100 blocks are written to the cache (using NAR data).  Then you have about 6.4MB of data in 15 minutes.  That's a bit over 51mb (6.4MB * 8).  You'd need to average about 60kbps over the WAN to fit in the 15 minute interval (51mb / 900seconds * 1024).

If you can't take a snapshot, then just look at the write bandwidth of the LUN (in the NAR) and use that number.

WAN Acceleration products like SilverPeak, Citrix WANScaler, etc can help reduce the bandwidth required and increase overall performance.  Replication over distance slows down as latency increases, the accelerators can improve this.

Keep in mind that MV/A over iSCSI is limited by the two active ports (one per SP) and the port bandwidth as well.  Probably not an issue but if you have lots of writes to replicate, then it could be a limiting factor.  Typically the WAN link is the slowest part of the link though.