Oveur posted a new blog on the state of the new server cluster. in brief the new monster "consists of 70 x IBM LS20 AMD Opteron Blades , each Blade sporting two 2.4 GHz 64-bit Opteron processors with 2-4 GB of RAM, depending on application. That's a 140 CPUs, a full replacement of the current 140 Intel XEON 2.8 GHz CPUs. All to run EVE."

yaaaaarrrdwaaaaaare aka new server cluster - yay!
reported by Oveur | 2006.01.17 10:41:47
I've mentioned before that we have bought a new server cluster for Tranquility, but I haven't gone into any detail. It's a long road of testing, testing and then more testing to be able to reach the point where we can order a new Tranquility - and then there is some more testing!

The main considerations this time around were if 64-bit processors would run our 32-bit code better than our current 32-bit 2.8 GHz XEON processors. This alone would be reason enough to order the upgrade, so we started testing a 64-bit proxy server with our 32-bit code. To our amazement, a dual processing 64-bit machine performed at a level equivalent to three of our current 32-bit dual processing machines. That was the point where we ordered, YAY \o/

But give me details ...

Yes, shortly. The main goal, however, is to use 64-bit code on 64-bit processors and the testing of this is still ongoing. Not only do we have to get Stackless Python and our own code running comfortably on 64-bit processors, we also have to make sure that our physics engine won't run into floating point problems when it's being used on different client platforms against our 64-bit server code (Called "Drift" or something like that).

Well, anyways, we ordered the new server cluster (YAY AGAIN) before Christmas, so it was kind of a big Christmas present for us here when we got confirmation of the order. It's now roughly a month since then and we have already started to recieve some of the hardware.

A recent extended downtime replaced a couple of the old proxies with new IBM 64-bit AMD processor blades, but the performance increase wasn't really noticeable in gameplay since the performance problems are mostly located on the solar system servers. After all, it's only 4 of 70 machines we have recieved.

Shut it and give me details ...

Ok, in a minute. We're not only upgrading the physical servers themselves, we're also upgrading the internal networks with new layer-3 and layer-7 switches between the individual Bladecenters. You see, instead of old Tranquility, which has all its solar system servers in 1U rack mounts, all with individual networking cards, the whole cluster is Bladecenters. That's 14 machines in each Bladecenter chassis. This allows us to put some really really big pipes between the Bladecenters, mostly multiple 1 GBit.

Not only that, we're also moving to a new hosting facility. Yes, that's right, the new Tranquility will be assembled and powered on with access to better internet backbones.

To begin with, we'll have similar network setups as we have on Tranquility today to try to decrease the unknown factors in the initial setup phase, but we'll then be in a position to connect to more backbones, further improving the connectivity to the new Tranquility.

You see, EVE doesn't use that much bandwidth at all. Its peak usage is about 50 Mbit on peak hours. We have "fat pipes" coming in, but what we're looking for is stable, low latency ("low ping"), well inter-connected backbones which provide the best routes to as many of our players as possible. This is achieved by getting high-capacity fiber connections to multiple internet backbone providers, which again should result in decreased network lag. (YAY \o/)

GIVE ME DETAILS!!111111

OK, I can take a hint. The new Tranquility consists of 70 x IBM LS20 AMD Opteron Blades , each Blade sporting two 2.4 GHz 64-bit Opteron processors with 2-4 GB of RAM, depending on application. That's a 140 CPUs, a full replacement of the current 140 Intel XEON 2.8 GHz CPUs. All to run EVE.

If the first tests are on the proxy servers and the performance increases shown there will apply to all the code we run, we should triple the processing power currently available on Tranquility. However, I personally am going to wait and see if this holds true - because it simply sounds too good. Then again, the tripling was done on 32-bit code; the gain could be even more with 64-bit code. Now isn't that a disturbing thought?

Oh, and had I mentioned that we're also planning on adding another RAMSAN ?. Yes, the awesome performance of this beast resulted in us taking one more :)

We're serious when we say we're addressing the lag , but the downside is that there is extensive testing, delivery schedules, installation and then more testing that needs to be done before it all materializes. We're spending serious money and resources on this because we hate the current situation just as much as you do. And the results of these efforts will materialize mid February. YAY \o/

Sales Merchant was kind enough to compile a list of serious questions about the new Tranquility cluster. They are as follows:

1. Does this mean the new super hamsters have arrived?
Yes, some of them have already arrived. We like to refer to them as "Yarrsters".

2. What will happen to the old hamsters?
They will spend their days watching gladiator movies.

3. Will the old hamsters recieve a company pension and lifetime's supply of cabbage?
We were fortunate enough to get Protein Delicacies for them. Apparently, there were some issues with the ingredients according to the supplier, so we got them cheap.

4. Will the new hamsters require daily downtimes to go to the bathroom?
Yes, we were unable to genetically remove that deficiency.

5. How long is the DT likely to be while the hamsters are installed in their wheels?
We require something like 24 hours. These are big and powerful Yarrsters and the wheels are quite elaborate. They also bite. And like gladiator movies. Fearful combination to say the least.

6. How spiffeh will Jita be on sunday evening at prime time?
Hopefully, we should be quite spiffeh. But you know what they say about systems like that. When there is room for more people, they will come to fill it up.

7. How many PVPers do you estimate will be able to blob each other to death in a single 0.0 system?
Enough to make the baby jesus cry.

8. Will you be changing your node setup?
No, we want be able to watch gladiator movies with the Yarrsters.

9. Will there be any server patch to complement the new hamsters?
Yes, we call it Blood.

10. How many concurrent users will the new TQ theoretically be able to hold?
Given enough Protein Delicacies, we should be able to double the current peak usage, but we would have to solve the Jitas and Renses of the brave new world first.

11. Can I have the old hamsters?
No, they will be busy watching gladiator movies, but I am in a position to get them on dates with you.

You may discuss this Dev Blog in this forum post .