I am running Windows Hyper-V Server 2012 R2 (Server Core) on 3 machines to test Hyper-V extended replication. The amount of data being replicated is significantly more than I anticipated. I also have some questions about the various options for
the frequency of replication, i.e., what impact that will have on replication performance.
Here is an overview of the hardware configuration.
HV1 and HV2 have an Intel Core i3-4130 CPU. HV3 has an Intel Core i5-2400S CPU. HV1 is setup using 7 HDD Storage Space with Double Parity. HV2 and HV3 have 6 HDD in RAID6 using a higher end LSI RAID controller.
HV1 and HV2 are on the same LAN with a Gigabit network. There is an OpenVPN tunnel between HV1 and HV3 which are in two different LAN networks. The upload speed of the link between HV1 and HV3 is 10 Mbps.
I am running a Windows Home Server 2011 VM on HV1. That VM is replicated to HV2 and then the replication is extended to HV3. The initial replication of the VM was done when all 3 HV machines were on the same Gigabit LAN segment. Afterwards HV3
was moved to a remote location.
WHS 2011 VM is setup with 3 VHDX files. The boot VHDX is a 160 Gb Fixed VHDX file. The data VHDX is a 4 Tb Dynamic VHDX. (My recollection is that only about 3 Tb of space is current allocated.) The WHS 2011 Pagefile is on a Fixed VHDX which
is not replicated. Only the Boot and Data VHDX files are replicated. Shadow Copies are disabled on WHS 2011 to eliminate any potential they have to add unnecessary disk changes that would need to be replicated.
I don't recall the exact terminology, but in case it makes a difference to performance ... I am using the Kerberos / Certificates method with HTTPS and compression to replicate the data. Rather than the other option for setting up replication (which
I believe required Active Directory?).
Q1) The amount of data to be replicated seems significantly larger than what I would expect to see.
Last night I copied 4 VHDX files to my PC. Three files were 8 Gb in size and one file was 10 Gb in size, i.e., 34 Gb total of new data to be replicated. I then started a backup from my PC to WHS 2011 VM. About 4-5 hours later I checked
on the status of the replication. The backup from the PC to WHS 2011 was complete. There was very little data waiting to be replicated from HV1 to HV2.
But the Extended Replication from HV2 to HV3 showed 98 Gb yet to be replicated. Given the initial change to the PC was only 34 Gb seeing nearly 3 times as much data set to be replicated between HV2 and HV3 was puzzling.
Q2) Would changing the frequncy of how often replications are done affect the amout of data that has to be replicated? Or does the amount of data stay the same and the frequency only determines how often Windows checks on whether or not data needs
to be replicated to the other machine?
At the moment I have the replication frequency on HV1 set at 30 minutes and on HV2 at 5 minutes. In the particular scenario described above I would have expected nearly all the disk changes to be "new sectors," i.e., I would not think that
the same "sector" might be updated several times?
Since the link between HV1 and HV3 is the slowest (which is used for the extended replication from HV2 to HV3) my thought was that allowing HV2 to check more often as to whether or not data needed to be replicated would help to keep the link utilization
high.
But just in case the same "sector" is updated multiple times, and if that should cause HV Replica to transmit the sector multiple times, my thought was that setting the frequency on HV1 to 30 minutes might lessen the amount of data that would eventually
have to be replicated from HV2 to HV3? For example, if sector 99 is updated 10 times during that 30 minutes I wasn't sure if 10 "instances" would need to be replicated? Or would only the last "instance" need to be replicated
to HV2?
Q3) Is there any difference to replication performance depending upon whether or not a Fixed or Dynamic VHDX is being replicated?
In my scenario if I were to compare a 4 Tb Fixed VHDX versus using a 4 Tb Dynamic VHDX (of which only 3 Tb is currently allocated) will using one or the other format almost always guarantee that more data will need to be replicated? For the initial
replication I'd guess that more work would have been involved for the Fixed VHDX than the Dynamic VHDX. But after that initial replication is done does the nature of using a Dynamic VHDX mean that there are more changes to the disk and thus more data
to be replicated? Or is there no specific correlation betwen the amount of data that would need to be replicated and the type of VHDX that is in use?
Thanks for any insight you can provide to any or all of these questions.
Theokrat