We need a stress test with layering off


(Daimonas) #1

Well, if this is how “no sharding and no phasing” works, what is the guarantee that “no layering” will work not the same way?

Essentially, the sharding and phasing engine is still up and running in Classic, it is just that the zones in particular have been merged, so they are all on one shard. But, as you can see, it is not explicitly set to “merge everything” - but, instead, they set it to “merge Elwynn, Westfall, Redridge, Searing Gorge, Burning Steppes, etc” - and forgot about one zone, that wasn’t so obvious.

Which means, that the sharding engine is up and running, and even if everything seems turned off, it may suddenly fire up during the AQ event, because of server load.

If it isn’t really possible to explicitly set “sharding off” on a server, then we need to be sure that nothing unexpected happens in the moment of truth.


#2

It would be nice to get at least some small info on how they are testing this stuff in the beta client. People are getting pushed around to different phases, seemingly at random (I know I have). Is that deliberate for testing? Are they still cleaning up how layering will work, and perhaps only gradually phasing out the sharding data from the client?


(Daimonas) #3

I don’t know, but I don’t like it. Don’t they have the technology to just hard turn-off sharding and cross-relam? I’m also talking about cross-realm whispers, that suspiciously work on classic beta even better than on retail, and about the “XXX-ClassicBetaPvP has joined the guild”


#4

I’m guessing that is what we had Wednesday night. All starting zone mobs camped and killed as soon as they spawned.


(Skribby) #5

The first stress test initial start was with no layers, they announced when the layering was starting. It ran fine, there were just soooo many people. It was kind of cool, but took me 30 min to kill 3 mobs.


(Daimonas) #6

you mean the May one?


#7

I would not doubt that there is a contingency in place for server and game stability that layering turns on if there is XXXX number of players in a zone. For example the AQ gates opening event, in the end player experience is more important than keeping to what happened 15 years ago. If anyone remembers being on a large to full server during the opening the server would go through a cycle where it was up for 10 minutes and during this time you would have 10-30 second ability delays, it would crash, be down for 5-10 minutes and repeat for the first 3-6 hours. This happened on 3 servers I had characters on US Laughing Skull, Shattered Hand(Shattered Server is more accurate in vanilla) and Illidan. In the end I would rather experience the event with a stable server over not playing because it is crashing. Give me Layering all day.


(Daimonas) #8

Honestly, I’d prefer login queues. As for the crashes - the event scripts itself caused a lot. Without the event, the realms were running fine, even at full population. So, technically, layering it would only increase the load, because every layer would be running them separately. And, also the amount of concurrent players would increase, because of the lack of login queues.


#9

Let me break this down. I will start this with I design data centers for major corporations and have deep understanding of how these should be designed( I have CCIE DC, multiple VMware and storage manufacturer certs). The crashes were caused by the high volume of players in one continent. Remember back in Vanilla each continent was it’s own server and the raids were located 1000’s of yards above Kalimdor in most cases. So this cause crashed and lag on client side since it was trying to display animations and details from 1000+ players at one time. Thus crashing clients and the server that was trying to run a one time event. Yes the realms ran fine when this same population was spread over 2 servers(each continent) and in different zones. That is the key to these issues. As for login queues, I knew multiple people that sat for 4 hours during the Shattered Hand opening event in a queue and missed 90% of the event, this is not enjoyable for anyone and is easily avoided by using modern technology called container programming. Container programming you say, I will give you the for dummies version. A container is a separate virtual machine that can be created and removed through automation requiring no intervention at all(this was not around 15 years ago and is what sharding and layering is based off of). So when you log on to a layer you are on a completely different layer than another player, thus increasing the resources for the server and in essence decreasing the overall load on one layers container. Since a character is only a reference in a database switch a character between layers is seemless as the orchestrator that sits in another container will just move the pointers to another layer/shard thus moving your character to a separate container. So, every layer decreases load, increases performance and therefore increases the end user experience.


(Daimonas) #10

Or you can make a container with more resources

Then it would also keep crashing like this in Dalaran on 3.2.


#11

This does not fix the client side issues, or the fact that distributing the load is significantly more efficient then boosting a single servers resources as you are also hitting network limitations, database limitations, and many other virtual limitations. Just accept that the best way to handle this from a hardware capability point of view is to distribute the load over multiple containers(layers) and that throwing more virtual space on a single server hits a dead end point where you lose capability of that hardware over having it split up. Put it this way having 4 separate servers with 1/4 the resources of one single sever with all of these resources is significantly more efficient and even the more important word of all of this is that it is significantly more SCALABLE. Scalable is most important part of container programming as it allows for layers to be commissioned and decommissioned seamlessly. This means when there is 4000 players in Silithus for the opening it can split this into manageable layers with no interaction from server admins. And as players leave it can decommission layers as needed automatically.

As for you Dalaran on 3.2 had phasing which was the start of splitting player bases into multiple servers but was poorly executed. There was also reasons that Blizzard avoided having actual events in major cities beyond end of tier boss handins. There was not large amounts abilities being spammed, etc. There is a reason there was only ever one AQ40 opening, it was an epic failure in execution and they learned from it. And all other events like this have been distributed over entire continents and for short periods in one location. And had a longer than 6 hour time to live.


#12

One last thing about you wanting queue’s. How would you feel about being in a queue for the entire AQ40 event and missing it? Are you OK with this?


(Daimonas) #13

but they already said they have a technology that lets the server work with thousands of people in one zone.


#14

That technology was sharding and now layering.