Contents migration upon S3 connection btwn our global servers

GIAI GIAI Internal
1

Have had done some studies on MinIO based S3 compatible storages. First of all, I appreciate @KeithLee to step up and run this server from Seoul, but I believe we are all benefited if we keep our own replica. With below rationale, I propose to sync all our servers across the globe.

We can certainly exploit South Korea’s cheap utility bills for electricity and bandwidth, but the physical distance still matter for my European servers. @catherinemaguire should be more concerned, once we run live video streaming on our own servers.

It is my understanding that Jitsi.Meet tries to pull client resources while BigBlueButton provides more dedicated solution from server resources. This nullifies our initial expectation for Jitsi being used for internal communications. Even for that, we will be benefited by BBB. Although Korean Bandwidth offers distinctive price advantage, unlikely Euro and American students experience nearly zero time lag. As such, I think we should have our own servers not only for S3, but also for BBB.

I am sure @KeithLee will build an error-proof S3/BBB server, but given our global scale services and our own research needs, we should have our own replica. Redundancy in servers is often needed for failover, but we need it for load balancing and regional backups.

For BBB installation, it comes to my understanding that we need to assign an uninterrupted public IP, so each of us need minimum two IPs, two MinIO servers (for local redundancy), and one DB replica. Even if we go with standalone Turn server, our dependency to public IP stands still. At least I know that @KeithLee has given up setting a coturn server behind his MicroTik router’s firewall. I also have given up handling UDP traffic with HA Proxy, in case you wonder why I insist on the public IP.

At the end of the day, most of us can still exploit our own heavy servers, and by combining DB, in addition to global pulling, we can use our resources more efficiently. At least, to my best knowledge, we do not need any extra machines. We just need to pay little more bandwidth and storage fees.

@ethanmcgowan and I might be able to share server resources, but I still think that we’d better keep our own. London to Zurich isnt short distance.

2

It really comes down to your choice, except the fact that for MinIO, it is better to set up the multiserver all inclusively, if @davidoneill 's proposal is accepted. Once the set up is done, as far as I know, it is not easy to sync remote servers. I can wait for a few days, if @catherinemaguire and @ethanmcgowan need time to think and prepare. I use 2TB x 4 NVMe disks on 4x4x4x4 board for a single node. I will place one in the primary web server and another on the secondary. It will be like cross server RAID 1. I havent done robustness check of this set up, so I am happy to have your suggestions and feedbacks of my set up.

Following will be the buckets to the S3 instance.

  • SIAI LMS videos
  • SIAI Website
  • SIAI Portal
  • GIAI Portal
  • GIAI Square
  • GIAI LMS
  • GIAI Research
  • MDSA
  • The Economy
  • The Economy Korea
  • OTT Ranking
  • The EduTimes

If EduTimes wants their own server, we should reconsider the last one. @catherinemaguire , plz have a little chat with them to see how much storage they need. We also need to double check if they want to split storage for MBA Ranking and Law Ranking.

3

Since not all buckets need to be replicated, I suggest to keep the sync at bucket level. You guys can set up an independent MinIO instance. As much as I don’t want to be affected by any of the global replicas, you guys probably don’t want to be affected by my MinIO configuration.

As for the initial sync, it is going to take some time to copy all files, so we should keep it at offpeak hours.

4

Sorry for the belated feedback. The EduTimes team took sometime to understand what we have been doing. They are ok to go with a single bucket for EduTimes/MBARanking/LawRanking, but was not sure if it is better to use our S3. Unless they use our NextCloud /S3 personally, I don’t think it will be a big deal. Newspapers hardly consume any storage. 100GB for them should be enough. @KeithLee Let me know if you plan to charge any cost to them for data upto 100GB, but given the integration btwn GIAI and EduTimes, I think 100GB for share wont be a big deal.

5

I really don’t think 100GB is a big deal, and given what our relationship, it is not something we should put a price tag. Unless they want us to assign 50GB per person, we’d better focus on what they would like to do with our server upgrade.

For instance, with our new Drupal website, its native multi-domain support, and internal S3, we can cover the same traffic at much cheaper price, the saving of which can be used for wider gloabl reach and more advanced functional support like big data dashboard.

Not sure what’s in their mind, but surely we can upgrade our MBA/Law ranking websites with more powerful functional support.

6

They prbbly will ask us to host EduTimes’s website as well in addition to MBA Ranking and Law Ranking, as the Drupal website can easily be extended for multiple domains.

I can see your Korean websites are already migrated to our central Drupal website. Not sure what order is in your mind, but once your team is done with GIAI and SIAI, I think we should re-approach them with EduTimes’s migration on top of sharing S3 storage. For now, 100GB sounds negligible, so it won’t affect much, but later, I kinda think they may ask us more, like access to NextCloud, BBB, and so on.

7

As @KeithLee mentioned, with Drupal, I don’t think we should worry about hosting another website both in terms of website control and server resources. 100GB is plenty for a newspaper hosting, unless they want to have videos hosted by us. Other than that, it is not a topic for discussion at this stage.

@KeithLee, let us know when you are done with full migration. Given what I see, both GIAI and SIAI websites can be migrated within a few weeks. It is only design work left, I assume. Setting up a federated S3 may take quite some time for all of us, so we should preplan beforehand. I guess Nov 30 / Dec 1 can be an option.

8

Alright. @davidoneill, I will set up Min.IO based in-house S3 in this weekend. I might need a day or two more for testing, but surely will be ready to help you guys set up each local copy in Nov 30 and Dec 1. I will create a doc in ERP for internal reference.

@catherinemaguire, you can do the testing for performand accesibility after we build this S3 behind Cloudflare’s load balancer. I hope the EduTimes team is satisfied with our new settings.