-
Join the MyBroadband community
South Africa’s biggest forum. Discuss, discover, and connect with thousands of members.
-
Question of the Day: What's the most interesting small town you've visited in South Africa?
Jumbo Frames
- Thread starter SBSP
- Start date
9KB Jumbos is is then! :twisted:
If anyone is interested.
Previously i meant 1500MTU not 5000MTU Minimum
There is a massive differance between 1500 MTU VS 9000 MTU.
Everyone who i spoke to dont even know what a Jumbo Frames are and say they just leave their NAS devices on the default network setting.
Incase anyone wonders, ist basically the chunks of data it send at a time between the source and target , anything over 1500 MTU is concidered to be a Jumo Frame. Like a HDD Block sizes when copying I.E Total Copy Vs Windows Copy.
Not recomended for Latancy sensitive apps but this works great for iSCSI.
Cheers
If anyone is interested.
Previously i meant 1500MTU not 5000MTU Minimum
There is a massive differance between 1500 MTU VS 9000 MTU.
Everyone who i spoke to dont even know what a Jumbo Frames are and say they just leave their NAS devices on the default network setting.
Incase anyone wonders, ist basically the chunks of data it send at a time between the source and target , anything over 1500 MTU is concidered to be a Jumo Frame. Like a HDD Block sizes when copying I.E Total Copy Vs Windows Copy.
Not recomended for Latancy sensitive apps but this works great for iSCSI.
Cheers
Last edited:
There is a couple of things to take into account.
1. What is your network setup?
2. Do all devices in the path from the QNAP to the end device accessing the data stores capable of handling jumbo frames?
3. Are you using fibre channel to connect to the LUN on the QNAP?
4. What data are you transferring between the two?
5. Do you know the average package size of the traffic being accessed?
1. What is your network setup?
2. Do all devices in the path from the QNAP to the end device accessing the data stores capable of handling jumbo frames?
3. Are you using fibre channel to connect to the LUN on the QNAP?
4. What data are you transferring between the two?
5. Do you know the average package size of the traffic being accessed?
Also thought Jumbo frames would make a big difference but everyone I spoke to said it all depends on your data on the SAN.
If you are running ESXI then it won't make much difference as you are not transferring big chunks of data. The important part is if it's turned on at one place it needs to be enabled everywhere. Ie: SAN, switch and server.
If you are running ESXI then it won't make much difference as you are not transferring big chunks of data. The important part is if it's turned on at one place it needs to be enabled everywhere. Ie: SAN, switch and server.
We configure 9014 as our frame size as far as Jumbo Frames are concerned, specifically on storage-related interfaces. This includes the iSCSI interfaces, as it was pointed out, as well as any and all Live Migration/vMotion interfaces.
You have to have JF configured on your SAN if you have virtual machines residing on it, purely because of the size of the data. If you don't you will have stipid high latency, and you will run into performance issues on your VMs.
EDIT> Remember that the same MTU needs to be configured on the network side to make things happen.
You have to have JF configured on your SAN if you have virtual machines residing on it, purely because of the size of the data. If you don't you will have stipid high latency, and you will run into performance issues on your VMs.
EDIT> Remember that the same MTU needs to be configured on the network side to make things happen.
I have only changed the MTU on the Qnap, and i'm using ethernet.
I will have a look on the ESXi Host and change acordingly. then Test again.
EDIT the ESXi is currently set to 1500MTU!!!!
Let me change it and see what happens
Ok i changed the MTU on the ESXi and its still running at ~1500 IO/s
But its OK, its better than ~140 IO/s when the QNAP is set to 1500MTU.
This server will be purley for capturing Video from 16 Network cameras at HD quality so i can imagine the VM will process allot of data.
Will see what hapens, if all fails i can always upgrade to a Fiber Channel.
But its OK, its better than ~140 IO/s when the QNAP is set to 1500MTU.
This server will be purley for capturing Video from 16 Network cameras at HD quality so i can imagine the VM will process allot of data.
Will see what hapens, if all fails i can always upgrade to a Fiber Channel.
Sorry, just a quick question, but what speed are the interfaces?
I made the assumption you had 10GbE interfaces, and in which case the JF would make a difference. Over a 1GbE interface you probably won't notice a massive difference, and in a lot of cases the the JF won't apply.
Sorry, that was an assumption on my part.
I made the assumption you had 10GbE interfaces, and in which case the JF would make a difference. Over a 1GbE interface you probably won't notice a massive difference, and in a lot of cases the the JF won't apply.
Sorry, that was an assumption on my part.
Last edited:
If you really want to fine-tune your storage network, I suggest see if you can find Yapt and run that on your nodes. This will gve you a good indication of attainable IOPS. You can have a look at IOMeter as well, but I find Yapt to work better.
its 1GB Interfaces and the tests i have done are shown in the second post of this thread using "IOmeter"
I first tested the IOP speed on C: of the VM of which the VMDK sits on the internal drive of the Host. (This was just to get a comparable output)
Then i did a test on
E: of which the VMDK resided on the iSCSI drive with the QNAP's interface set to 1500MTU and it gave me 147IO/s after changing to 9000MTU i got over a 1000 IO/s.
Since the VMDK will be written to constantly with Video data i suppose 9000MTU will be about fine for large amounts of data being written all the time.
Since it will be a video cam server it will be writing to iSCSI drive all the time
I first tested the IOP speed on C: of the VM of which the VMDK sits on the internal drive of the Host. (This was just to get a comparable output)
Then i did a test on
E: of which the VMDK resided on the iSCSI drive with the QNAP's interface set to 1500MTU and it gave me 147IO/s after changing to 9000MTU i got over a 1000 IO/s.
Since the VMDK will be written to constantly with Video data i suppose 9000MTU will be about fine for large amounts of data being written all the time.
Since it will be a video cam server it will be writing to iSCSI drive all the time
Thats the next step, I dont know how to yet, I'm using an SMC POE Switch, which is managable.
Just have to figure out how to go about it.
[EDIT] If i dont come right i'm thinking to just connect the QNAP directly to the 2nd Nic of the ESXi Host