Questions About Azure Instance Allocated Bandwidth

Question

I know few facts from Azure:

1. Limited bandwidth for each instance, for example: Extra Small instance has 5Mb/s
2. We only pay outbound bandwidth, inbound is free
3. For same data center, traffic is free
4. even for small instance, I checked the network interface, the connection speed is 10G. So Allocated Bandwidth is always lower than the connection speed which the network interface capable with.

I like to know:

1. Allocated Bandwidth for outbound only or for all traffic?
2. Allocated Bandwidth for outside of data center or for any data?

I like to know more for useful answers to me:

1. Will the Allocated Bandwidth limit the traffic between instances in one deploy (they will be in same data center, same network) I will exchange data between instance, so I like to know if the whole network interface capacity will be utilized.
2. Will the Allocated Bandwidth limit the traffic between instance and CloudDrive which in same data center? If does, then limit on Read or Write or Both? I will use CloudDrive lots, and the bandwidth won't cost since they are in same center, so I like to know the speed limitation.
3. Will the Allocated Bandwidth limit the instance connection to outside, for example, to send email by a outside SMTP server?

Any official source is highly appreciated.

Answer 1

At the end of the day, you are mostly constrained to the NIC on the box from within the datacenter. Each of the physical machines has a 1Gbps NIC on it. It was found to have about 800Mbps sustained transfer speed. Since each host could also currently have 8 guests on it, the ratio of the reserved NIC was a multiplier of the cores. If you had a Small instance, you were reserved 100Mbps, Medium has 200Mbps, L 400Mbps, etc. However, this does not taken into account bursting. Small instances can burst to 250Mbps or so in practice - it just depends on what your neighbors on the box are doing. The more cores you reserve, the higher your bursting will be upto the max.

The XS instance size is actually limited to 5Mbps, so it does not follow the other pattern.

Again, your connection within the datacenter will mostly be limited by the NIC bandwidth. This is true for instance to instance and instance to storage. For out of datacenter (where prevalent network conditions generally matter more), the NIC is the limit still, but generally other factors outside of datacenter are bottleneck.

The sort of exception to all this is when accessing a particular partition in storage: You can get around 60MB/s transfer per blob (or partition really), as this is limited to the rate at which a partition server will service your request as opposed to NIC speed. However, the entire storage account's limit is 3Gbps (more than your NIC). That can only be achieved when accessing multiple partitions.

Back into your context, that means you will get at max around 60MB/sec (480Mbps) for a particular cloud drive. To fully saturate that, you would need a Large instance or higher. That is why I said you are mostly constrained to NIC.

Answer 2

Regarding your questions,

• bandwidth throttling is applied to outbound traffic only.
• it is for any data, even if the VM talks to another VM in the same cluster.
• If you have multiple instances in one deployment, bandwidth throttling will apply to the traffic between them.

You may see How Does Bandwidth Throttling Work in Azure? for more technical details.