I apologize, I'm going to hijack this thread somewhat - but I wanted to go over something Gregg just stated, something that has been a bane of my existence for more than 13 years...
[ QUOTE ]
The exadata has a crazy amount of cache. We can host most of our database in the cache. So we are esentially getting the benefits of SSD combined with really fast disk performance.
[/ QUOTE ]
The number of times I've heard this argument. Its the argument that hardware companies make when they want to sell something more profitable....
I have a Seagate "hybrid" Momentus in my mac - I just fitted it the other week. Its a 32Gb RAM and 7200 RPM 500Gb drive. According to seagate, I "essentially get the benefits of SSD combined with really fast disk performance".
Now, in real world testing - I'm glad I made the change for my personal use. My Mac is running a little faster - but predominantly because I've gone from a 200Gb 7200rpm drive to a 500gb 7200rpm drive.
But is it really as fast as SSD? I mean, c'mon - putting the cache directly on the disk and ensuring it buffers up the data of frequently read data - surely thats going to be within a fraction of a percent of a real SSD.
Anandtech did a comparison between the hybrid HD, a regular drive and a true SSD. The results are here :
http://www.anandtech.com/show/3734/seagates-momentus-xt-review-finally-a-good-hybrid-hdd
Now - if you look carefully, you'll see that the SSD wins pretty much hands-down EVERY time. Certainly the hybrid is faster than a regular drive - but when you compare the factors between SSD and the hybrid, its a pretty huge gap. Especially on write performance.
As Jeremy stated, " There is just something that warms my heart about measuring disk latency on micro-seconds instead of milliseconds". I'd back that up by stating "I'd rather see disk WRITES in nano-seconds rather than milliseconds". Thats the crux that Jeremy was getting at. With JDE, the reads and writes are very, very substantial. JDE talks to the database in a very "chatty" way. Certainly, read performance is one factor - and it certainly SEEMS that its the biggest factor due to the fact that JDE reads more than 90% more than its writes. However, if a WRITE takes too long, then the READ, which is sequential, is going to queue up until that write finishes. Therefore, even though the reads are happening faster, the spread of WRITES occurring against the database are going to dramatically affect the performance of that database.
I published a whitepaper back in 1999 while I was at JDE (its available on my website). We evaluated an SSD on our standard test. Back then, SSD's were $30,000 per Gb at least - so to get one was VERY cool. All we did was place the archive logs onto the SSD - and JDE improved performance by at least 15% instantly.
That was without any tuning. It was a huge eye-opener. For more than 13 years, I've been stating that the incorporation of SSD's in the database architecture will occur. Today, with the new Exadata2 from Oracle - I think that clearly shows the importance of SSD in architecture design.
So, certainly while your Exadata has a lot of cache - its not the same as the Exadata 2. The SSD's make a huge difference.
As for my Mac - well, I'm quite happy having spent $200 and improving performance and disk space. I could have spent $2000 on a true 500Gb SSD - but in MY CASE, I didn't want to spend that kind of money on an older computer. My plan is to purchase a new Macbook Pro (when the quad cores come out) - and at that point go 100% SSD.
Just my 2c. I'll let the thread go back to the original discussion of VMWare now....(!)
