For the past few months I've had the Sun X4150 on a long term test, and I've been throwing a wide range of different applications and problems at it to see what breaks.
As I've mentioned before, my test unit has two quad core Xeon CPUs, 16GB of RAM and four 76GB SAS drives, and it saps about the same amount of power as a T1000, that is, in terms of the electrical power required to run the machine. But it is, in fact, a significantly more capable machine in some respects, albeit for a different target audience and range of applications.
I won't go into the list of operating systems supported on this machine - as an Intel-based box the world is pretty much your oyster.
The box is very powerful for what is a single U high pizza box. The eight cores and potential for up to 64GB of RAM make this a small powerhouse of computing, even before you include the storage that you could fit into the same space. There's also the Integrated Lights Out Management (ILOM) system and four Gigabit ethernet ports. Oh, and two PSUs so that you have network, drive and power redundancy if you want it.
It may seem like a lot to squeeze into such a small box, and believe me, when you look inside, you won't believe that there actually seems to be some space inside there. No, really, it's not a lot of room, and, to be fair, a lot of that space is required to keep the air flowing through the box, but you still feel that it is space that could be better used.
There are lots of tests that have been run on the box, and although it's tempting to quote many of the synthetic benchmarks that are available, what you really want is the real-world examples.
It's taken some time to get the various kinks and problems and optimizations worked out on the box. These are a constant, but ever present annoyance.
What I was really looking to compare was the power usage of the box between the T1000, which is a low power, low-MHz, but many threaded box against the multiple cores of the X4150 in terms of the response times and performance. I used the usual combination of the web applications (blog, Wiki and Cheffy to test the performance).
The difference in performance of the X4150 is evident once you start running the tests. The X4150 managed almost 2800 req/s on the basic image tests, approaching a 20% increase on the T1000. For the blog tests, I managed an average of 242.12 req/s on the X4150, compared to 197.87 on the T1000. That's another 22% increase on the T1000. All this is with the same basic power consumption.
The 8 cores are working hard, but the 32 cores on the T1000 are keeping up. Where you do see a difference is when you start to ramp up the the number of simultaneous requests. The T1000 can handle many more requests (although the response time starts to extend slightly) before you start to get rejected items. With the X4150, even with the faster CPUs begin to reject requests much sooner once the simultaneous requests get above a certain level.
The blog test is read-only, and probably a significant part of the increase can attributed to the built-in RAID. In a write test (with random data into a Wiki), the single drive on the T1000 proved much more of a limiting factor, with a 37% increase on the X4150, almost certainly because the multiple drives and the RAID made enough of a difference.
Another test, this time comparing ZFS (RAID-Z) to RAID-5 showed a small benefit in favor of the hardware, but by less than 5% in real terms. This is impressive, both for the hardware, and the efficacy of RAID-5.
I'm still tying up the different tests, and occasionally still see a potential optimization, but for a 1U high box, the X4150 is decidedly capable.
I'll go back to some other aspects of the hardware and environment for the X4150 later in the week.