This blog post was originally posted on November 10, 2008. It was then withdrawn the next day and replaced with the short explanation below. Since the person who asked me to withdraw it has since left Sun, I have left Sun, and Sun itself will not exist anymore in a few days, I have decided to re-publish the original text for historical purposes.
The background to the original blog post was the tensions we faced inside Sun, where a majority of revenues still came from pushing Sparc servers, even if the future supposedly was in being a great open source software company. Sun had launched their new multithreaded CMT Sparc servers and discontinued their old Sparc line. The bad news was that MySQL 5.0 performed very poorly on those heavily threaded chips. On the other hand Sun account managers were busy trying to sell the expensive Sparc servers instead of x86 servers that MySQL consultants would recommend you to use. This had lead to the situation where every week I would have to explain to a customer why this server that is twice as expensive, is half as slow as their previous Intel servers. (And then you wonder why Sun was in financial trouble...)
People working in Sun's system division had already published some "benchmarks" and blogs supposedly promoting MySQL to be used on Sparc systems. This gave independent MySQL consultants (like Percona) easy ammunition to accuse Sun of giving bad advice to customers. To then launch a full on marketing campaign based on outright falsehoods seemed like a path that would lead to even more trouble.
The motivation for my below blog post was simply to make a statement that we, the actual MySQL experts, did not support the marketing message coming from Sun's Systems division. I felt I was a good person to do it, since I knew people with more responsibility in the MySQL division would not be allowed to speak up against Sun's interest of desperately pushing their hardware with the MySQL brand bolted on. I never thought lying as part of your marketing was a good strategy anyway, but it certainly isn't a good strategy when specifically targeting the open source community! Or I don't know, maybe I'm just allergic to lies... I was actually told it is more acceptable in the USA than in Scandinavia. True?
When I was asked to remove the post I immediately did so, since the statement had already been noticed by the MySQL community, which was all that I wanted to signal anyway. And now I've restored it for historical purposes, since by Sun being part of Oracle this topic doesn't matter anyway. (Btw, I've seen benchmarks with Oracle RAC running on these Sparc machines: It seems to be a great match!)
The following text was here from November 11, 2008 to January 24, 2010.
Out of loyalty to someone I respect, I have chosen to withdraw my post about honesty that appeared at this address.
Let me however take this opportunity to remind my dear MySQL customers and users: I warmly recommend you consult an experienced MySQL expert when architecting your MySQL solutions. Setup a call with a Sales Engineer, hire a consultant, read the blogs of MySQL AB oldtimers. If you have a hard time finding solid MySQL info, we are there to help you.
This is the original blog post, published November 10, 2008 to January 24, 2010.
I remember when I was interviewed for the current job I have as MySQL Sales Engineer, one thing that was explained to me was that it was important to be honest with the customer and be clear on where we recommend MySQL (and even more this practice is true for selling MySQL Cluster) and where we don't. This left a good impression on me. After all, with an engineering background I have to admit I was used to thinking of salespeople as some lowlife scum that you can never trust. So to be introduced to an attitude of being clear and honest certainly made it easier for myself to start working in sales.
Speaking of being honest... Let me start by upfront apologizing for the idiotic stuff someone in Sun marketing is putting out as part of the otherwise laudable effort to launch new server configurations tailored for various MySQL tasks. Let's look at the worst example first:
I found all of this via Kaj's blog originally. Hmm... "Cross-platform MySQL Runs Best On Sun" ...with such headlines, maybe Kaj should watch out not to be drinking too much of the corporate koolaid himself! In particular my attention was caught to the claim: "Virtualization. Get up to 3x higher throughput, 4x power and space efficiencies, and 10x price/performance on MySQL deployments."
That's certainly weird, I know virtualization is the new buzzword, but for the life of me I cannot imagine how it could significantly improve performance! On the contrary, I know it may often decrease it! So I followed a few more links to learn more. The claim comes from a real customer case DigiTar. We learn that:
- * 4.5X performance improvement over HP DL145 G2 servers
- * 10X price-performance improvement
- * 2X increase in processing capacity
- * 75% reduction in operating costs for MySQL applications
- * 83% reduction in power consumption and heat, representing a 5% decrease in total power consumption
- * Consolidated 8 HP servers to 2 T2000s
- * Consolidated 8 load-balancing switches to 2 Sun N1216 switches
So this is where Sun marketing makes such misleading claims that it could in all fairness be considered lying. Let's go through that point by point.
- Yes they got 4.5x performance (throughput) when comparing the newest Sun T2xxx servers to a 3 year old HP model. This what you'd expect from Moore's law (double performance in 18 months, paraphrased). So the good news is that Sun T2000 servers probably would've performed quite nicely even against modern HP servers. But unfortunately Sun Marketing did not choose to honest about this.
- 10x price performance is now nonsense since it essentially buils on the above 4.5x figure which is nonsense. Even so, it seems the additional 5.5x is mostly "because we love Sun so much" kind of TCO calculation, not based on anything very factual.
- 2x increase in processing capacity. Hmm... What happened to 4.5x times? What does this mean? Reading the story, seems like the real increase in MySQL processing power is only 2x and the 4.5x is something else, like website throughput. If that is the case, then a 2x improvement when upgrading from 3 year old servers is actually quite poor...
- 75% reduction in operating costs: Not elaborated in story.
- 83% reduction in power and heat: So this is the big marketing punch line for CMT, let's give them this one. (And sure, using less electricity is great.)
- Consolidated 8 HP servers to 2 Sun T2000: Sounds good.
- Consolidated 8 load balancing switches to 2 from Sun: Sounds good too.
The fact that T2000 is not a good fit for raw MySQL performance is well known. There are good uses for those servers, even for running MySQL (consolidation). In fact, my collague Ivan Zoratti made a good presentation on this at the European MySQL conferences, I hope he would post that on his blog so I could link to it. There are many nice things Sun Marketing could write about T2000, instead of lies like this that risk spoiling also the reputation of not only Sun but MySQL.
And by the way, nowhere was virtualization mentioned? And yes, you can run multiple instances of MySQL on one server without virtualization, so for any of the above to work you don't need virtualization. No, you don't get 3x better throughput by turning on virtualization.
The thing to remember is that Sun is a big company and many kinds of people work there. You may sometimes hear good and bad things the same day. Don't come and blame me, if someone else is saying something studip.
Just to give an opposite example, check out Some MySQL technical videos from Jennifer Glore, especially the last video. This is Sun Principal Engineer Allan Packer talking about MySQL Performance. Clear numbers, clear presentation, the truth and nothing but the truth. In fact, as a sales person I almost get the temptation to say: Hey Allan, how about a little less honesty for a while, at least in public? Just kidding, this is how we want it :-) Facts, not marketing.
The good news of course is that Allan's team is working on it and many performance improvements are already there for MySQL 5.1. If you want me to name the most important benefit of Sun acquiring MySQL, Allan is it!