Slashdot Mirror


MariaDB CEO Accuses Large Cloud Vendors of Strip-Mining Open Source (zdnet.com)

Big cloud companies are "strip-mining open-source technologies and companies," complains Michael Howard, CEO of MariaDB. At their developer conference, Howard accused "big cloud" of "really abusing the license and privilege [of open source], by not giving back to the community." ZDNet reports: Even as MariaDB grows by leaps and bounds in enterprise computing at Oracle's expense, Howard sees Oracle and Amazon fighting against it. "Oracle as the example of on-premise lock-in and Amazon being the example of cloud lock-in. You could interchange the names, you can honestly say now that Amazon should just be called Oracle Prime...."

In the first keynote, Austin Rutherford, MariaDB's VP of Customer Success, showed the result of a HammerDB benchmark on AWS EC2... In these tests, AWS's default MariaDB instances did poorly, while AWS homebrew Aurora, which is built on top of MySQL, consistently beat them. The top-performing database management system of all was MariaDB Managed Services on AWS. "My first reaction when I looked at the benchmarks," said Howard, was "maybe there's incompetence going on. Maybe they just don't know how to optimize a DBMS." He observed that one MariaDB customer, one of the biggest retail drug companies in the world, had told MariaDB that "Amazon offers the most vanilla MariaDB around. There's nothing enterprise about it. We could just install MariaDB from source on EC2 and do as well."

He then "began to wonder, Is there something that they're deliberately crippling?" Howard wouldn't go so far as to say AWS is consciously doing a poor job of implementing its MariaDB instances. Howard did say, "And then it became clear that, however, you want to articulate this, there is something not kosher happening." Howard doesn't have much against AWS promoting its own brands... But, if AWS's going out of its way to make a rival service look inferior to its own, well, Howard's not happy about that.

ZDNet adds that "it's also quite possible that unoptimized generic MariaDB instance will simply lag behind AWS-optimized Aurora.

"That said, even in this most innocent take on the benchmark results, cloud customers would be wise to take into consideration that cloud instances of any specific software service may not be created equal."

6 of 200 comments (clear)

  1. Aurora performance is awful, too by shatteredsilicon · · Score: 5, Interesting

    I've consistently seen Aurora MySQL performance worse than RDS MySQL on writes, and overall, a carefully tuned (in a way you can't tune RDS or Aurora due to lack of access) MariaDB instance on EC2 will utterly annihilate Aurora and RDS on overall performance, for same instance sizes. Of course, a bare metal setup of the same size (same CPU cores, same RAM amount) will annihilate the EC2 instance on performance and at about half of the TCO over three years, but nobody seems to care about that these days.

    1. Re:Aurora performance is awful, too by whoever57 · · Score: 3, Interesting

      Fear of not being able to scale seems like the #1 concern in SV these days,

      As well as scaling issues, what moving to EC2, Google Cloud or other solutions gets you is a possible reduction in system admin costs.

      Of course, there is a minimum: if you only have one system admin, you probably can't reduce your system admin costs.

      --
      The real "Libtards" are the Libertarians!
    2. Re:Aurora performance is awful, too by sfcat · · Score: 3, Interesting

      Fear of not being able to scale seems like the #1 concern in SV these days, but I swear to you for the average startup there are many, many more important problems you have to solve first.

      Exactly. And even if they don't have scaling problems, since all the *cool* kids are working on scale issues so should they. At company I worked for recently, one that you have probably heard of and probably has your data, at peak they handle about ~12000 web requests a minute...not a second, a minute. That's 200/second, on a 50 server array. That's 4 requests a second per machine...4...not 4000, or 400...as in my phone could do it without me noticing a slowdown in performance...4. So of course they spent gobs of money moving from their custom and very modern data centers with high uptime to GCE with terrible performance, just so they could "scale". Remember, 4 requests/machine/second is the max they had to handle before. Maybe 10% of the 600 engineers they had understood just how silly this all was. But the CTO, who started as a PHP programmer at the company when it was first starting, didn't understand any of this and just let his "professional" managers run everything into the ground. Its absurd really....the software industry is now a parody of itself...

      --
      "Those that start by burning books, will end by burning men."
    3. Re:Aurora performance is awful, too by phantomfive · · Score: 3, Interesting

      Recently I saw a company with three different nosql databases (and a relational database for slower things). This was to support fewer than 1,000 users. I showed them the Amazon "How to scale to 10million users" video (it's all over youtube), and sent them a summary of the bullet points (at around 1,000 users, we should consider getting off sqlite3. By 500,000 users, we need to have performance monitoring in place). That gave the non-technical people a road-map idea of priorities, and linking to Amazon gave it authority, even if they didn't actually watch it.

      --
      "First they came for the slanderers and i said nothing."
  2. Re:Open source by ConfusedVorlon · · Score: 4, Interesting

    That's hardly fair; For example; GPL is open source, but explicitly designed so that people who use it are under certain circumstances required to contribute back to it.

    The way the GPL is written; Cloud providers like Amazon don't have to contribute back to the project - but that's probably not what people wanted when they came up with the GPL.

    e.g. it's probably a legal bug, not a legal feature

    there are a bunch of different licences, I'm just using GPL as an example here.

  3. Amazon's RDS storage is intentionally crippled. by rMortyH · · Score: 4, Interesting

    Amazon's RDS offerings are really convenient, from the perspective of making snapshots and setting up replication. But, I have never been able to push their default 'SSD' storage past 60MB/s. (PostgreSQL and Mysql) That's terrible. That's less than USB 2, and even some SD cards can do that! Our on-prem can do 180MB/s on spinning rust and around 550MB/s on (obsolete) SATA SSD. If you want anything better on RDS you have to REALLY pay a premium for IOPs and transfer, or pay a premium for way more ram and a ton of caching, in addition to external caching in the rest of your stack. I have not used Aurora on RDS, so I don't have a comparison, but I have my suspicions. It would be pretty easy to just give you a few more MB/s and make it look a whole lot better. Luckily in our case we could optimize things enough that storage performance didn't matter too much. But RDS storage performance is so pitiful that it's seriously worth considering putting your DB on a bare metal box somewhere with NVME storage and just put up with the network latency and get 50 times the storage performance (and more ram and cpu while you're at it ) at a fraction of the price.