Questions on cdp proxies tcp connections

[FrenchBlue]

Hello,

I'd like to understand how tcp connections work between source and destination proxies, to be able to optimize our architecture. We are experiencing low cdp throughput in general, with associated issues (VMs remaining for too long in CBT mode during high I/O activity, initial sync of large VMs taking ages, etc...).
I've done some iperf3 tests between source and cdp proxies (on very distant data centers), and single connection numbers are quite low, probably limited by latency, in the 20 Mbits/s range only (local network is full VMware 10 Gb on both sides, properly configured, VMs use vmxnet3 etc..). However it scales very well, ie running 10 connections in iperf3 in parallel gives about 10 times the throughput of 1 connection.

So I'd like to know how it works exactly between cdp proxies. Do they use 1 connection per cdp job or 1 per replicated VM? Or completely something else?

Thanks.

[veremin]

One connection is opened between the source and target proxy for each virtual disk processed. Can you test the connection on HTTP and HTTPS ports (80 and 443)? We have seen cases where performance varies depending on the port. Thanks!

[FrenchBlue]

Hello,

Thanks. To do this I think I have to disable Veeam services on the target proxy if I want to use iperf3 on these ports? I can try. Will also try the same iperf3 tests between Linux VMs on both data centers.
Meanwhile, I checked the network traffic of the source and dest cdp proxy VMs, and it's not stable, does a sort of regular sinewave with a period around 2 minutes and the lowest part near zero, unfortunately I can't attach a screenshot here. That's weird.

[FrenchBlue]

So, after quite a lot of investigations, I ended up increasing the disk cache size on source and destination cdp proxies from 50GB to 100GB. We don't have more than 50 vdisks replicated at the moment (recommendation is 1GB per vdisk) , but after this, the network throughput between both proxies went up by about 15x from average 40 Mbits/s to 6-700 Mbits/s, and stable instead of the sort of sine wave before. However, since I also rebooted the proxies, I'm not 100% sure the cache size was the issue, it's Windows... If it occurs again, I'll try to just reboot the proxies after having disabled the cdp jobs. I'd rather have cdp Linux proxies too, but not available.. yet?
And BTW, iperf3 on Windows is not reliable, only measurements with iperf3 from Linux VMs are relevant.

[veremin]

I'd rather have cdp Linux proxies too, but not available.. yet?

Not yet, but v12 will make you happy. Thanks!

[FrenchBlue]

Hello,

Good news for the cdp proxies, thanks. For the other initial issue here, it worked fine for a few hours only, then traffic between proxies has dropped to almost zero and all VMs have gone to CBT mode. I'll open a support case asap.

[veremin]

It seems like a connection issue (the communication between proxy servers got broken, resulting in the source proxy server switching to CBT mode).

Kindly post the ticket number here once you get one.

Thanks!