I have bought a CRS309-1G-8S+IN for upgrade my lan network to 10Gb.
I have connected my wmware server with to dac sfp+ 10Gb cable , my qnap nas with an S+RJ10 module and my computer in 2.5Gb with also an S+RJ10 module.
i have also an aruba 1930 connected with dac cable aruba for the the equipements which don’t need 10G.
when my vmware server is connected on the aruba, i have average 250MB in wrting and 600MB in reading
When i connect the vmware server on the crs309, i have 250MB in writing but only 15MB in reading.
With my computer connected in 2.5GB , i have 250MB in writing/reading.
i’m a noob with routerOS and i don’t undersand why i have so poor reading performance with my vmware server.
Most likely, the traffic is processed by the CPU instead of the hardware, and that is the reason for slow networking. Post your interface and ip config, please, so we can analyze it.
Your configuration looks pretty straightforward, i.e. no routing, just switching, which is good. What speeds are the ports reporting? Do they show up as 10G? Are all the devices in the same LAN subnet?
yes all the devices are in the same subnet.
In the doubt i will check again this evening but it seem the ports show up as 10G.
As you say, my configuration is straightforward, i just want 10G switching nothing more!
Hello,
i assign the ip on the bridge but no change.
I just noticed that when i do a iperf between my computer (connected in 2.5G on the crs 309) and the crs309, i have 2.5Gbit in upload but only 250Mbit in download.
if i do an internet speed test, i have also 250Mbit upload and 600Mbit in upload (600Mbit is the max of my internet contract).
Finally the problem is not only with my vmware server…
That drives me crazy
Did you try to enable flow control (both Tx and Rx) on all involved ports?
The thing that bothers a switch the most is speed change - from faster to slower port. In this case switch has to buffer data and we all know that buffer bloat is bad. So when that tiny buffer fills up, switch has two choices: 1) drop next frame or 2) signal transmitter that it has to pause sending … the later is done by deploying flow control. The former is default behaviour, but hurts TCP throughput imensely … because in this case transmitter has to re-transmit and with relatively high-latency (milliseconds on 10Gbps networks is high latency) this kills throughput.
I see you have VLAN filtering turned on, maybe VLANs are not configured properly. Would you be able to share a full config export and removing any sensitive info (public IPs, passphrases, usernames, serial numbers, etc)
Post #11 above (by @jericho63) IMO shows that throughput problems are not due to traffic hitting CPU (it’d be much higher than 2% at 2.5Gbps) but some other reasons, internal to how switch chip handles the traffic.
High CPU load while running btest is usual but has nothing with normal traffic handling (which is port-to-port, not port-to-CPU as in case of btest).
Does it work correctly if you use 10Gbit on both connections? (or do you only have asymmetric connections).
Sometimes flow control helps but you might need to go to the switch port settings and set an ingress/egress rate limit to see if you can get closer to the actual hardware line rate.
One way to identify this is if an upload from 2.5 GB connection is line speed (to a 10GB connection) but download is under 1GB then this is most likely the issue.
@mkx
i have try to enable flow control following your reply.
For the test (compturer to crs with btest) i have enabled it only for the port where is connected my computer but not better, i have always a througput in download arround 250-300Mbits with btest and an througput at 2.5Gbits in upload, internet speedtest 195Mbits in download / 595Mbits un upload.
If i try a test with my nas (with nas performance tester 1.7https://www.808.dk/?code-csharp-nas-performance) , connected on the crs309 as my computer, i have 211MBytes in wrtting and 269MBytes in reading.
@darthgizm0
i have disbled valn filtering since a while but no change.
during a test with btest in receive mode, profile show arround 20% on network and 20% on btest.
@biomesh
I have vmware server and a qnap nas connected in 10G and my computer in 2.5G.
But i have also througput issues with my virtuals machines to internet or to the NAS.
I will try an upload from my computer to one vm and see the througput.
The best way to test is using iperf on both ends. As long as the machines on both ends are powerful enough to generate/receive 10Gbit worth of traffic, you won’t hit any other bottlenecks like disk IO etc.
i know i will have a bottleneck in reason of the disk i/o limit on the vmware server because they are not ssd but for the moment i have a througput below it.. and it is very frustrating.
i will perform new tests this evening.