Good news, techsters!
The 2U server has been delivered, yesterday afternoon, 1 day earlier than my most recent estimate. (Specifications at the bottom of this post, after all the blah-blah-blah.)
This means that I can run memtest86 (opteron version) overnight tonight, and hand it over to the developers tomorrow.
Here is my schedule for today -
1. At 9AM I have an interview with Voice of America. That should take only 10-15 minutes.
2. After that I will be going to my office downtown to pick up various this and that (raid card for 1U server, Fedora Core 1 and 2 ISOs etc.) and heading to the colocation facility.
3. When I get to the colocation facility, I will do whatever I need to do to get 3 of the 4 1U servers online immediately -- the Wikipedia seems half dead to me this morning, I have a feeling the Yahoo feed has launched and we're getting slaughtered. Hopefully those machines can be pressed into service almost immediately to help relieve things.
4. For the "new zwinger", I will need to do whatever I need to do to get the 2x250gb RAID array with fedora core installed. I think this is going to be easy, as I already have 2 hard drives in a mirror with Fedora core all ready -- but I made these in an old machine of mine, and nothing ever works as planned with computers. :-) But once I get that going, I'll hand that one over as well.
5. The people at Silicon Mechanics gave me a full explanation of this, and after we discussed it, they went ahead and shipped the new server with *32 bit* Fedora Core 1. That's because *64* bit Fedora Core 1 would not install. But Fedora Core 2 (64 bit) *would* install, and they would have been happy to do it for me, but because they have a slow isdn line, and didn't have the ISOs yet, it was going to take them longer than it would take me, so I had them just ship it.
So this means that I will be installing Fedora Core 2 64bit.
Additionally, they set up the RAID 10 like this: 219GB HW RAID 10 (split over 2 Channels)
But James Day has convinced me that this isn't exactly what we want. I'm going to review our discussions of that, as well as discussions on the list, and then I'll also show up in IRC to make a final confirmation of what we want to do.
-------
Here are the specs...
Dual AMD Opteron 248 - 2.2Ghz - 1MB L2 Cache Memory: 8GB (8x1GB) PC2700 Registered ECC HDD (6): Seagate Cheetah 15K.3 73GB U320 15KRPM SCA SCSI SCSI Controller: LSI MegaRaid SCSI 320-2 - 2 Channel U320 RAID 64MB Cache w/BBU CD-ROM Floppy 460W hotswap redundant power supplies (to be place on two different circuits!)
--Jimbo
On Thu, 27 May 2004, Jimmy Wales wrote:
- When I get to the colocation facility, I will do whatever I need
to do to get 3 of the 4 1U servers online immediately -- the Wikipedia
I'll assume somewhere in there you'll make some attempts to get SATA DMA enabled as per our earlier discussions off list? I tried both those kernels on my Dell Diminsion 8300 which has a similar ICH5 although the Dell BIOS doesn't support putting the controller in "legacy" mode. Both see the PATA and SATA drives just fine w/ DMA. The console log is in the same place as the kernels.
I would not recommend using either of those kernels for anything but tests as I didn't have enough information to know exactly what needs to be enabled (eg. I might not have built the right network driver module...)
they have a slow isdn line
I feel their pain (and moved to a cable modem.)
Here are the specs...
...
Do you know what MB is in the system? Is the memory configuration NUMA -- each CPU has it's own memory pool? (This is a important thing to know and should be noted for building an optimized kernel. There is a slight performance penalty for "off cpu" memory access.)
--Ricky
On Fri, 28 May 2004 02:07, Ricky Beam wrote:
On Thu, 27 May 2004, Jimmy Wales wrote:
- When I get to the colocation facility, I will do whatever I need
to do to get 3 of the 4 1U servers online immediately -- the Wikipedia
I'll assume somewhere in there you'll make some attempts to get SATA DMA enabled as per our earlier discussions off list? I tried both those kernels on my Dell Diminsion 8300 which has a similar ICH5 although the Dell BIOS doesn't support putting the controller in "legacy" mode. Both see the PATA and SATA drives just fine w/ DMA. The console log is in the same place as the kernels.
Do you know what MB is in the system? Is the memory configuration NUMA --
Jimbo is in the colo now, and by visual inspection, the mobo is: on the 4 new P4s: Superdot P4SCE, Rev. 2.1 on the older P4s: Superdot P4SCE, same as above, except Rev. 2.01
each CPU has it's own memory pool? (This is a important thing to know and should be noted for building an optimized kernel. There is a slight performance penalty for "off cpu" memory access.)
--Ricky
Wikitech-l mailing list Wikitech-l@Wikipedia.org http://mail.wikipedia.org/mailman/listinfo/wikitech-l
On Fri, 28 May 2004, Jerome Jamnicky wrote:
Do you know what MB is in the system? Is the memory configuration NUMA --
Jimbo is in the colo now, and by visual inspection, the mobo is: on the 4 new P4s: Superdot P4SCE, Rev. 2.1 on the older P4s: Superdot P4SCE, same as above, except Rev. 2.01
SuperMicro... super with a filled in "O". That's a nice little board. Heh, "Optional CPU overclocking"
http://www.supermicro.com/products/motherboard/P4/E7210/P4SCE.cfm
These have a hardware watchdog that could be of great value (esp. to whomever has to drive over to the colo to reset a hung box.)
On the BIOS Advanced Settings->BIOS Features screen, enable APIC mode if available. It'll improve IRQ routing. An SMP kernel will be needed to use this capability.
On the BIOS Advanced Settings->Advanced Chipset Control, "On-Chip Serial ATA" might need to be set to "Enhanced Mode". Auto might also work. "SATA Mode" should be "IDE" (RAID is intel's software raid. linux doesn't support it at the moment.) Set the S-ATA port modes to SATA Master. In theory, that should get the SATA drives off by themselves at full SATA-150 speeds.
--Ricky
Ricky Beam wrote:
These have a hardware watchdog that could be of great value (esp. to whomever has to drive over to the colo to reset a hung box.)
Fortunately, unlike our old colo in San Diego, the Tampa facility is staffed 24x7, and so merely resetting a hung box is a matter of a phone call. I will be adding a few more people to the authorized list before I head for Europe, but right now it's Jason and me.
--Jimbo
I am new to wikipedia and I have notice that all article can be joined by an url like : http://*.wikipedia.org/wiki/words_in_the_subject_of_the_page
I suppose this choice better than http://*.wikipedia.org/wiki/article.php?id=12343234234 was to have easier url and better indexed urls However I am surprised, the choice of "_" ( underscore ) in the url is not as good as "-" (dash) or even "." ( dot ) in the url
could the developer rewrite the url function ? this would lead to much more traffic comming from google
see that : http://www.markcarey.com/googleguy-says/archives/ \ underscores-are-not-word-seperators-in-google.html
I have tested it and that's true !
For those who haven't been living on IRC the last day...
New machines yongle, maurus, rabanus, and perhaps will have been added to the apache cluster. If DNS isn't quite updated: 207.142.131.237 yongle yongle.wikimedia.org 207.142.131.238 maurus maurus.wikimedia.org 207.142.131.239 rabanus rabanus.wikimedia.org 207.142.131.243 will will.wikimedia.org
We apparently now know how to get DMA working on those SATA drives, thanks the combination of a BIOS tweak and Ricky Beam's magical kernel concoctions.
Tim's updated the dormant parser cache in 1.3 with memcached support, which will give us a great way to fill up the memory on those new machines with pre-parsed HTML and save a lot of unnecessary re-rendering.
Whee!
-- brion vibber (brion @ pobox.com)
Brion Vibber wrote:
We apparently now know how to get DMA working on those SATA drives, thanks the combination of a BIOS tweak and Ricky Beam's magical kernel concoctions.
Should we make it a priority when I get to the colo today to get DMA working on all of the other 1U boxes? It's important because I will be out of town for a month, so if we don't do it now, we will not be able to do it until much later.
(Actually, Michael could come in and help with it, but that's a bit more difficult since he's not familiar with the setup as much as I am.)
We need someone onsite to do it, because of the required BIOS tweak.
--Jimbo
wikitech-l@lists.wikimedia.org