Recommended Posts

It's interesting, since changing my ESXi server and linux guest to use MTUs of 9000 and vmxnet3... (Basically, I've got a bot running) the latency and replies that the bot is taking to respond previously were about half a second... Now it's between 4-12 seconds per reply, I'm really not sure why the performance has gone terrible :s but it really has noticably gone to rubbish.

Does your switch support jumbo? Does the physical nic support jumbo, did you set it to jumbo?

You sure are not going to be able to use jumbo out to the public net, so if your connections are mostly to the public net - jumbo frames are not going to get you anything.

Jumbo needs to be end to end to be worth it. If do not have end to end support for jumbo - it can make matters worse, yes.

You can view your physical nics settings on your esxi host using.

post-14624-0-19940100-1335803481_thumb.j

Now I would have to assume when you change the vswitch to 9000 mtu that this would change the physical nic as well, but you might want to verify it is set with the above command.

And what switch is your physical nic connected too - does it support jumbo?? Unless it is at least a smart switch might have issues. Also what level of jumbo does it support. Problem with jumbo can be that different makers use different sizes 9014, etc.. etc..

looking forward to the nano binary - that would be sweet!

As to vswitch setup, there isn't much too it.. You have not done any advanced option changes have you?

One thing I would double check.. You mentioned you changed your MTU to 9000 at one point. From your last screen of your vswitch you show 1500. Did you remember to change the MTU/Jumbo Frame setting on the nic in your vms?

have you done any changes to the nic settings in the vms?

Mine are default settings, the only thing I do on my vms related to network settings is I do disable ipv6 if not going to be using it. I turn off the llmnr (link-layer topology discovery mapper/responder) crap

IPv6 is fully disabled, not just uncheck the binding - on the vm

reg add hklm\system\currentcontrolset\services\tcpip6\parameters /v DisabledComponents /t REG_DWORD /d 255

This for one makes my ipconfig /all much easier to read ;)

Other than that I don't recall doing anything else. So we are on the same hardware, same os even for vms -- WTF your vswitch only showing 300mbps with iperf while mine shows 1.55Gbps?? You are showing a 10gbps connection on the vm nic now that you changed to vmxnet3 right?

At a loss here to what could be so different in our setups? Same freaking hardware N40L, esxi 5 u1 from vmware. Are you running your esxi off a usb disk or anything? Mine is installed to the 250GB drive that came with the N40L

Have made any changes to the bios settings of the N40L? What bios version do you have - mine is dated 7/29/11 I believe - I know its 7/11 something.

"I still need to test the thick/thin disc theory yet though." - this might have some issues with your file copies to and from vms, but should have nothing to do with the speed he is seeing on his vswitch - have you run an Iperf test between vms, and to box outside your host. Do you also see bad performance? Cpressland was seeing like 355Mbps, I show like 800 between vm and box outside. And 1.55Gbps between vms connected to the same vswitch. Iperf would remove any disk bottlenecks from the test.

Have you taken sniff using one of your vms -- is it possible that your vswitch is swamped with other traffic, maybe a broadcast storm or something going on? Turn on all the networking items -- do you see anything weird.. Mine is pretty much zero -- not doing anything currently.

post-14624-0-81550800-1335540080_thumb.j

Have you done any tweaking of the power management setting? Mine is set to balanced.

Under your security profile - what services do you have running/stopped?

post-14624-0-17908000-1335540711.jpg

I have EXACTLY the same set up and I don't see any performance issues! I have also installed my ESXi on the Local 250GB disk that ships with it.

I was running 3 or 4 VM's The CPU was hardly touched. The memory was at 7GB out of 8GB

Granted compiling the Kernel on my Ubuntu VM was slow but hehe cant complain about that its not like it was a production machine!

So chuck - what kind of performance are you getting pulling and pushing files to your VMs from physical boxes on your network. And what kind of performance are you seeing between vms just using the vswitch?

Are you also seeing 80+MBps to and from vm's in file copies, and are you also seeing well over 1GBps between vm's with iperf?

I'm with you not seeing any issues in performance - I am more than pleased with the performance for such a budget setup. I could not ask for more in this price range. The little box just screams -- the only thing that is a bit of question is why putting files directly on the datastore is so slow. But this seems to be an issue going way back, and not really anything to do with the hardware. If you look into this -- people have been complaining of the slow copy speeds to the datastore for quite some time on all kinds of hardware, even enterprise grade. But to be honest this is not something that requires lots of use -- you put a iso on there, your upgrades/patches to esxi, etc. It's a now and then sort of thing - not something your doing daily. I can live with 5MBps when add a OS iso I am going to use for install, etc.

Dude you got other problems then -- I just got on there with both firefox and chrome and had no issues starting a thread.

Here I started it for you

http://communities.vmware.com/thread/400609?tstart=0

Dude you got other problems then -- I just got on there with both firefox and chrome and had no issues starting a thread.

Here I started it for you

http://communities.v...400609?tstart=0

Thanks dude, not entirely sure why it doesn't work. Maybe it's my work connection breaking something....

Does your work run through a proxy? Does it do filtering? If you saying css is broke in multiple browsers I would assume the css page is not loading for some reason.

They have quite a few different css pages that are loading

post-14624-0-03354300-1336049971_thumb.j

Where I have seen issues with css -- github use to cause me lots of grief when I had rc4 ciphers disabled.. I had to go back to allowing rc4, too many sites ssl/tls is just broke to be honest.. Why sites are not all using tls 1.2 by now I have no idea.. Shoot unless they recently added with with the 12 update, I don't think firefox even supports it yet. **** its been around for like 3 years now, we all know there are issues with tls 1, but support and use of 1.2 is not really there yet..

Man that got off into left field quick ;)

Sorry about the posts one after another but time passes and you can't edit existing posts.

I setup a FreeNAS VM with a device extent of a thick disk. 6GB of RAM and no other VM's running. Nothing running over the NIC on my PC either.

Check out these uber upload rates over iSCSI on a 1Gig link (On Board NIC).

*Edit*

7.10MB/Second now and still slowing down :(

Patch doesn't fix startup btw.

*Edit*

Just found a thread that mentions about enabling write cache in the BIOS. Changing this actually improved my performance although its still poor. My NIC was showing 7% usage before it's now showing around 14%. That's about 18.7MB/Second now according to Windows. Seeing about 40MB/Second reading from the drive.

post-75681-0-20042200-1336078105.jpg

This topic is now closed to further replies.