-- RobynLanders - 07 Aug 2008

MOVED TO MFCF MEDIAWIKI

This page has been moved to the MFCF wiki as of September 2010. Don't bother editing it here.

MC3015 Machine Room Renovations, ST#60768

There's lots on the go in renovating MC3015, such as air conditioning, electrical service, new UPSes, networking, etc. The part I'll start off describing here is the relocation of existing equipment. There are quite a few ST items created already for this, linked together to some extent to show dependencies about which things need to move where before others move, and so on. The plans so far are incomplete and subject to change as we rethink how things will proceed, and are complicated enough to warrant describing and collaborating on, which is why it seemed like a good thing to put in a wiki.

Rationale

What's the point of doing all this?

Part of it was the expansion of the machine room into space formerly occupied by Operations. I don't know the specific reasons why that expansion was done. (One recently observed problem with the expanded area is that the electrical sub-panel has only 100 Amps of capacity, which is not much.)

A major goal is to disassemble the lengthwise row from in front of the air conditioners. This has several benefits. Airflow and cooling efficiency should improve by setting up proper hot and cold aisles. CSCF can add to the rows in the space that they have been allocated, and MFCF can add to its rows; both were impeded by the lengthwise row. The air conditioners become unobstructed to facilitate servicing or removal.

While reorganizing, the guidelines include trying to keep similar categories of machines together, trying to standardize on certain types of racks and keep them together, and trying to get rid of ancient racks. This is in part a cosmetic issue.

Row 10: Research Row (Middle Atlantic)

Looks like we'll turn row 10 into the departmental research row. The end state currently anticipated is this, going from the wall outwards:

network rack | Stats Rack | C&O rack | AM rack | RFU | Apple cluster

These are all Middle Atlantic racks.

Current state: Done!

Stats Rack has been move here from the lengthwise row in front of an air conditioner. The HP Teaching Rack was moved here from J7 as a transitional step and has since moved out again to yet another temporary location in January 2009. Winisk also moved out of this row in January 2009. In February 2009, the Applied Math rack was reloaded into a rack with casters, and the Apple cluster was moved here. There is space in this row for another rack (that's the RFU).

Row 14: Sun Row

We won't put a telco rack at the end because the number of network feeds for this row is going to be pretty small (the IQC rack has its own switch already). In the longer run, PDS2 needs to be replaced with a new electrical panel (and transformer?). The new IQC equipment has arrived; this is their collection of new Sun servers which occupies an entire Sun rack. Their current servers (iqc1.math and iqc2, both SGI Origin 3200s) will be retired.

Current state: Done!

PDS2 | small space | kea (cpu117) | Stats Sun Rack | UPS for Stats Sun Rack | new IQC | IQC UPS
UPS for kea

Two open shelving racks and a storage cabinet were moved out of this row. The kea/cpu117 rack is in place. Its UPS is movable but not very far; due to anticpated replacement of PDS2 and required cooling space on the sides of the UPS, the UPS is not in the row, it's in the space between two rows. The Apple cluster moved to row 10 in Feb 2009. The Stats Sun rack (biostat, transom, iqfi, redington) and its UPS moved here in March 2009. A UPS for the IQC rack has arrived (April 09) and we await electrical connection. Temporary power (no UPS) has been provided for everything except their blade server.

Row 17: Networking Row

There's already a large wiring rack in place against the wall. The current rack containing Extreme Switch 1 has been cleared out and core network stuff has moved here from the nearby block of small racks beside flexor, and those old network racks were removed, clearing the way for relocation of flexor.

Flexor

Flexor has been turned 90 degrees so it matches the crosswise orientation of the other rows instead of going lengthwise. This frees up space for another crosswise row near the transformer.

Winisk, IQC SGI rack

With the relocation of flexor, space became available to move winisk from row 10 to newly opened row 27 in January 2009. Winisk has been expanded and has its own full rack plus an accessory rack for UPS and storage. It may yet move again, to the new area. The IQC rack of SGI equipment was moved beside winisk in Feb 2009, clearing room in front of an air conditioner beside pilatus, where we expect to temporarily located kazan.math (K. Lamb's Altix 3700).

New Area

In the new area we could orient the racks lengthwise down the room in two long rows, or crosswise like the old area in three shorter rows. If we use crosswise short rows, we match the orientation of the rest of the room, we make use of the windows for visibility to all the rows, we have room for 18 or 19 racks, and we have room for one or two set-up benches and two or one storage cabinets. If we use lengthwise long rows, we contradict the orientation of the rest of the room, we provide window exposure only to the front row, we have room for at most 17 racks (realistically more like 15), and we have room for one set-up bench and maybe one storage cabinet.

It would seem then to be best to use three crosswise short rows.

SGI row

In the newly expanded area, the row closest to the ramp would be a short row of only 4 or 5 racks. No plans yet for what to put there.

The next row, in the middle of the area, could be an SGI row:

network rack | winisk UPS | winisk rack | kazan.math racks 1, 2 | pilatus racks 1, 2, 3 | pilatus UPS

Note that pilatus rack 3 and winisk UPS do not exist yet but are in the purchase planning stages; the kazan.math racks arrived in Jan 2009. So this would be an all SGI Altix row. Lots of new electrical work needed. Also, winisk will have 3 shelves of disk storage with it that need a home. The SGI racks will all be 100% full. So I'm hoping we can use some space in the network rack for that. (I recall Tom mentioning that he ordered dual-post network racks for this sort of reason -- in case we need to put some stuff other than just network switches in them.)

Alternatively, if the third pilatus rack never materializes, there might be enough room alongside the reoriented flexor to move pilatus there. Then the short row could be the SGI row, like this:

network rack (with winisk's disk arrays) | winisk UPS | winisk | kazan racks 1, 2 |

HP row

The last row closest to Operations would be HP racks for whatever. We've talked about moving the MFCF infrastructure machines here. The undergraduate teaching equipment has been remounted in an HP rack and would thus join this row.

What About Everything Else?

Yes, there's plenty of other stuff that needs to be reorganized and moved, and no, it isn't planned yet. That's what this document is intended to help accomplish.

Can we move everything at once and avoid intermediate steps that cause some things to get moved more than once? Unfortunately, no. We would have to shut too much stuff down for too long, we'd have to plan and deploy too much electrical reorganization, and so on. So we'll do it incrementally, trying to minimize repetition as much as possible, and watching for opportunities to move things at convenient times (e.g. end of terms).

Topic revision: r14 - 2012-09-06 - BillInce
 
This site is powered by the TWiki collaboration platform Powered by PerlCopyright © 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback