We are looking at performing an upgrade from 6.5 to 7.1 with a hardware change. We are looking at using containers on a T2 processor. I reviewed the support materials and they only say SPARC, I’m assuming since the T series is an UltraSparc its supported.
The questions:
Is anyone running on a T series processor?
If so…
Are you running 6 or 7 on the T series?
What is your impression of the performance of the T series? Do you notice a speed increase because of the parallel thread processing?
We are looking at performing an upgrade from 6.5 to 7.1 with a hardware >change. We are looking at using containers on a T2 processor. I reviewed the >support materials and they only say SPARC, I’m assuming since the T series >is an UltraSparc its supported.
The questions:
Is anyone running on a T series processor?
Yes, we are running both Niagara 2 on a T5120
and a couple of T2000 machines
If so…
Are you running 6 or 7 on the T series?
Running 6.5, first install of 7.1.x is just done in test though.
What is your impression of the performance of the T series? Do you notice a >speed increase because of the parallel thread processing?
We have no performance issues whatsoever. The machines have more than enough under the hood to run our entire environment.
Did you have any particular “case” in mind?
Is anyone running under Solaris 10 containers?
Yes, we are doing that too. There are some issues though, especially when it comes to native threads that we don’t have a good resolution for yet.
And the patching is a bitch, at least if you’re running many zones. (Patching of the OS and the resulting downtime on the IS/Brokers that is)
What is your impression of the performance of the T series… We have no performance issues whatsoever. The machines have more than enough under the hood to run our entire environment. Did you have any particular “case” in mind?
[COLOR=royalblue]No particular case, I should have been more specific, we are thing about going from an UltraSPARC III+ to the T2. My guess would be that we should see a performance increate because of the threading model in the T2.
[/COLOR]
Is anyone running under Solaris 10 containers? Yes, we are doing that too. There are some issues though, especially when it comes to native threads that we don’t have a good resolution for yet. And the patching is a bitch, at least if you’re running many zones. (Patching of the OS and the resulting downtime on the IS/Brokers that is)
Do you have a forum thread that talks about these issues or would you care to elaborate at all? I could see how OS patching could be a pain. Are you running multiple environments (dev/QA/Prod) on a single machine in containers? Are your machines clustered at all? We are being pitched a solution where we would have QA and Dev zones on a single machine but I don’t think that will fly because of the OS patching.
I think you will just find that you can run many many more processes paralell because of it.
The single processes though, does not use paralell threading very well IMO. But at least you get something to grow in [/COLOR]
Well no, actually. We’ve just been working with this internally and I think on a SR to webMethods.
The problem is what we think, the lwps-max-setting in Solaris containers set to 2.15G, which seems to be, not enough. Though you cannot mess with another container and this value if one of your containers hits the roof, which is neat.
And well, no. We’re not using different environments on the same machine. One for prod, one for ver and so on.
To be honest, we have dev & test on the same physical machines. Though verification and production can use their own iron :-]
And we have two physical for each environment, with a couple of zones on each. (two zones for IS, depending on who is using them. Then a couple of broker zones, though we are taking some of them offline now) The zones running the Integration servers are local zones and then we have IS clusters. Though the brokers are running on HA-zones that can move between the datacenters. (This only sucks during patchtime of course, since you can’t move a HA-zone to a machine with older patch level, and you have to patch it somewhere)
Well, just ask away if you’re wondering anything more. :-]
When the Sun T2000s were released we did a performance comparison between our current platform, a Sun V240 (twin UltraSPARC IIIi 1.5 GHz) and a T2000 (T1 8 core 1 GHz). This was about 2 years ago.
At the time we were running IS 6.5 SP1 on JVM 1.4. (I didn’t do any Broker tests.)
The test I ran on both platforms was to open a flat file from the filesystem, transform it to XML, then convert it back to the same flat file format and write it out to the filesystem. I used 5k, 66k, 150k, and 480k sized flat files, and also varied sequential vs parallel vs different numbers of parallel. (For those interested, I used JMeter to trigger the services and capture results.)
Generally across the tests the T2000 performed better than the V240 - with some exceptions. As the file size increased the performance difference narrowed. In the case of the largest file, at sequential or two processing in parallel the V240 was better (more GHz). At 4 in parallel the platforms were even, at 10 the T2000 was better.
This shows that 2 cores at 1.5 GHz does better at a few big jobs than 8 cores at 1 GHz. And 8 cores at 1 GHz handles lots of little jobs better, but slows down as more grunt is required (transformation of increasingly larger files). Which gels with what swl stated.
THEN… it was recommended I try the 1.5 JVM (which at this point wasn’t supported by wM).
I re-ran the same tests. With the 5k and 480k files average processing time was cut by a third, and with the 66k and 150k files average processing time was halved.
Given that this question is about the T2… a quick reading of the spec.org SPECjbb2005 benchmarks points to the T2 having a little over double the performance of the T1 with regard to Java processing - I’d like to re-run the same tests again on a T2 to get some hard figures.
Last weekend we did a hardware upgrade from the Sun V240 to a Sun T6320. This was due to forthcoming load over Christmas expected to exceed current capacity and one of our Business Units informing us that they had an agressive client role out plan over the next few months. All too much for the current box.
