About the talk
Do you know how multiple large game publishers around the world successfully launched hits without any scalability issues using a relational database? One thing that everyone knows in the industry is how painful and costly scaling a relational database is.
Dive into Google Cloud’s gaming database solutions with a deeper focus on the relational, yet scalable, database Cloud Spanner. Built on top of the internally battle-tested Spanner, it can support tens of millions of queries per second for the most demanding games and will help you build future-proof games that will grow with your players.
Speakers: Sophie Hu, Takayasu Oyama
Google Cloud Next ’20: OnAir → https://goo.gle/next2020
Subscribe to the GCP Channel → https://goo.gle/GCP
product: Cloud Firestore, Cloud Spanner, Cloud SQL; fullname: Sophie Hu;
event: Google Cloud Next 2020; re_ty: Publish;
I'm Sophia The Specialist today, I'm going to talk about how class Banner companies application experience. Come out and help you predict many things today, and it's amazing. But one thing, it cannot predict is a successful game. So what are we start? Davis solution, unpredictability compute, When it gets a lot of players, you're really the bottling because like my Sequel and not the alignment to scale horizontally. Funny gaming companies, solve the problem.
The more you must leave the more players your game. Last one is far from shorting where you all separate players population. What is a game back? And that's where Clause Banner Cloth Banner is the first Enterprise Street Houston. Database server is the cloud relational. Database structure was non-relational horizontal scale to Patient assistance and transactions across row shirts. Cost of relational data customers were using more power. General awesome. Keep their
relational data. Also serving traffic with zero downtime. Respond. Let's take a look at some examples of companies using for a surprise heat with Pokemon go to scale 2/50 X case scenario which reduces their operational cost. Another example is Dragon Ball from Uncle is classified as a base, reduce the risk related to human R Us as well as operational until spanner. And some of the benefits of Cloud 9 to the Google Network offers better speed and security or other clubs.
Stitches together for services. Has two instances Regional or synchronous replicas. The Regent. Incarceration locations close to or how does it help? Showtime is a way to synchronize cloth, clocks on all machines. Across clocks to p.m. end of time for vacation. Different notes can make decisions about the otter. Option section operation made from other nations to generate were executed went really fast. Global Network Showtime, external possible, which cable is broken down into several
Call spinner servers serve each replica. Reality of sweets, according to the amount of data. NASA's office. Splits is replicated most responsible for managing actions for bass Elite. Operation for the leadership, roles amount of clocks. Let's talk about what skin looks like. It's per second per second. Has been determined that we need glasses for a total potential, stress showed off, 6000 rights per second per second, Instrument of our best estimates determined that
spanner should be sufficient for the number of know shortly after lunch specials, 40 solvent leads per second per second. Restarting the day off for some time before Master returns to research lab rules and we find ourself with significant additional cost with a site to add additional note. Describe some advantages of spanner. I'm so excited. Thank you for my name is soccer in a software engineer. Barbie, and please allow me to do it recently. Dragon Quest walk is part of a long-running dragonfly series owned by a company.
The role-playing game where you walk around and fight with your monster in a real-world location using GPS and a r. Are some numbers to give you an Insight of how much data we fought this every day. The game was released in September of 2019, in Japan. How many millions of monthly active users by over 10,000 Todd on vacation? Ruby's Pub in processed thousands of cheering for second. Why we chose. Then I will talk about how we integrated into our development. After that, I don't know. How do you Center in real life?
Mario. There's reason and civility. The first one. well, for a while but when the numbers of guard, Going in and out with Jason, and we scaled every month whenever we released and then we wouldn't want the traffic. But we were able to do this with one place. The second reason is called. Like I mentioned in the previous slide used to spend days getting my people, bracing for another place. I need the code to buy and difficult to maintain. These things made a costly to add features and fix them. Allowed us to simplify the code. So adding features and fixing bugs to be done a lot cheaper.
I miss my Sofia, daughter has a lease or 9 that's always when we stopped at a constant rate. You haven't had any of these parts. You been running this game for 10 months now and we haven't had any major downtime. Now, I would like to talk about how we have you can see how to replace my people for storing to submit data. You might have noticed that my people still use that later. When you first decided to integrate planner into the game. Different operations to me.
Well, that was good driver framework. What's the driver was written by replacing my people driver with a driver. I said I was done. You join the game development team. Am I doing these preparation? The games about my team was able to fully integrate Center in about 2 weeks. Now, I'd like to give you some item on the list of access to share data. Volume. Now, if we access the data in Spanish, we would get what's called a hotspot. This happens because all the io is happening on one slice of data, that cannot be distributed.
But there is a way to work around the way to handle. This is how he acts like one of the Stars around them. In a doctor, the query and I will distribute the load. So we decided to use a bow and arrows where there is a gradual, go unload. This is not what we expected to have. Now, what we did? So, we did a simulation and plunged a project that has massive traffic. It's an adjustment period to allow the Lord to trigger and increase the number of queries rather. Wait until there was enough to have no desire to get it too many, if you would
starve would slow down wedding process. The right amount of cruise. I didn't do what. I've just described there are two things, you should be aware of when doing a warm-up. The number one is, Do not drop the tables after the warm-up. You would want to delete the me records that were inserted during the warm-up. Don't do it because that will, that will allow the 48 hours. So make sure you warm up, your database 48 hours prior to the last thing I want is based on how much traffic work so I can
get more turkey. When we Still In, by reducing the number of nodes to aggressively, you noticed that the latency in like 30 seconds. Three, we resulted to timeout in 3 seconds is the latency seems to be proportional to the number of nosy. So the takeaway here is don't try to reduce the number of nodes in one go, but we use them gradually in small interval. That's all I have for you today. Thank you for your time and get my keys with you. Thank you, Chaka, for your great story for the Nexus staff. We have a few more
Buy this talk
Buy this video
With ConferenceCast.tv, you get access to our library of the world's best conference talks.