/i/2004829164.png?fit=%2C&ssl=1)
You are proper, I did imply ‘Rare’ entry.
I do not know how a lot energy some previous SAN options in numerous knowledge facilities mix plus the additional community elements price. And whether or not that outweighs the cloud storage prices (~€0.24 per GB per yr for energetic storage incl. 3 to 9 replicas, for ‘Rare’ that’s ~€0.12).
____
Certainly, I assume that GitLab within the Twitter publish with ‘Object Storage’ refers to Google Object Storage.
After all it would not should be with Google Cloud, it can be accomplished with AWS S3 (Object Storage) and AWS CloudFront (CDN) and S3 Insurance policies (rights) and Object Lifecycle Administration. Together with AWS Lambda Features that may execute requests (identical to Google Operate) based mostly on the lifecycle occasion and append the cache-control header to recordsdata based mostly on replace/put occasions.
And naturally CloudFlare can be used as an alternative of AWS or Google CDN. Even perhaps the most cost effective model of CloudFlare is nice sufficient as a CDN characteristic on this setup.
Optionally, the CloudFlare serverless capabilities can facilitate numerous Git capabilities (on AWS that is doable with Lambda@Edge within the N. Virginia area), however I do not assume that’s vital.
____
The benefit of Cloud Object Storage is the multi-zone storage (a number of duplicates in a number of knowledge facilities), a reliability with 11-nines, excessive uptime and excessive parallel community connection to your storage.
In follow, for instance, I’ve examples of greater than 8,000 concurrent requests on a newly uploaded file of 1 GB that are dealt with with out issues. Effectively inside a minute the greater than 8,000 servers are able to learn and interpret the information. A median velocity of over 133Gbps.
Whereas all different learn and write actions additionally proceed undisturbed.
That’s with the AWS s3 customary storage class with out CDN. This can be totally different for a decrease storage class.
___
To sum it up, to illustrate you spend $1 million per yr in storage prices (solely) on the most cost effective model of FireStore ($0.20 per GB monthly = ~400 TB) and also you migrate that to Normal class Object Storage ($0.02 per GB monthly).
The saving is then:
$1.000.000/($0.20 – $0.02) = $900.000,-
With Google CDN you pay for cache fill per GB $0.01 and never for the lifetime of the cache.
Egress for CDN is $0.01 per GB and for Object Storage a minimal of $0.03.
That signifies that a CDN on prime of the Object Storage is already cheaper if solely half of the recordsdata are requested earlier than updating. That appears fairly regular for a Git repo. So in all circumstances it’s good to mix CDN with Object Storage for Git.
The FireStore egress prices are equal to the CDN egress prices.
Which means that the CDN fill prices are extra prices to the present prices, so have to be subtracted from the $900,000 financial savings (as a result of they’re extra prices).
That signifies that with 75 petabytes of GIT push knowledge monthly (900PB per yr), object storage with CDN is now not cheaper.
Notice that that is all based mostly on record pricing. I assume that within the above scenario, the Google Account group will definitely supply a big low cost.
____
It’s due to this fact cost-effective to not decode the textual content recordsdata in a Git push (with content-encoding gzip), however merely go away gzip compressed and retailer them within the object storage and land them in CDN as properly. For instance, the storage, fill and egress prices are decrease.
[Reactie gewijzigd door djwice op 5 augustus 2022 20:40]
