Apps built on the same public cloud are faster ?

[connieandcharlesw]

if two SaaS applications (one deployed on GCP and the other on AWS) are communicating then everyone just assumes there will be latency issues.

I am failing to understand, without knowing SaaS applications tenants how could we determine cross cloud communication will introduce latency ..
E.g
1 SaaS A is on GCP in US east and SaaS B is in Australia on GCP
2 SaaS A is on GCP in US east and SaaS B is in US East on AWS

How would someone determine in the above scenario #1 is preferable over #2 ??

 

[clovispower]

@connieandcharlesw I mean just think in geographic and physical terms. Data travel to Australia by virtue of just the distance is going to have higher latency than two data centers communicating on the same continent.

Whether or not your application can handle that latency is a design consideration, but not typically a design consideration you need to worry about at the starting up scale. Multi region, multi cloud, multi continent are considerations for much larger companies to be making.

 

[connieandcharlesw]

@clovispower Made a correction on my use case #2 if data is moving between the same geo why people oppose #2 so much ?? … couldn’t agree more. In the initial phase I haven’t seen a startup who starts with a multi cloud design .. (and also become successful)

 

[clovispower]

@connieandcharlesw In the case you're communicating across two clouds there are extra network routing hops involved, those hops introduce additional latency compared to intra datacenter communication and additionally for the intra datacenter case things are usually communicating on much higher speed interfaces further bringing down the latency.

 

[odem]

@connieandcharlesw Electrons travel at a certain speed. If the distance between two points is greater then it will take longer.

 

[notaperfectchristian]

@connieandcharlesw In general yes they will always be faster as there's probably less networking gear involved. However is that enough to make a difference will probably vary. It's old data but you can see there are many AWS GCP regions that have sub 10ms latency. source

However what you also have to keep in mind is that a single region isn't just one data centre, in the context of AWS, not even a single availability zone is guaranteed to be one data centre. Thus you might be crossing more network boundaries than you expect.

The cloud providers actually do work together a fair amount to ensure good peering connections between them (and the ISPs). It works in their favour to play nice because then more people will multi cloud, than to try keep people just within their cloud.

So really the question is, for the backend how much latency is tolerable? There also may be other advantages such as being able to have integrations for your customers that are more efficient because they're in the same cloud+region as the customer.

That said, multi cloud single region is more complicated than single cloud multi region, and multi region is more complicated than single region, so unless there are very large advantages to the former options you're unlikely to see anything but single region in a newer SaaS.