Hardware ordered October 2004

From Meta, a Wikimedia project coordination wiki

In general we anticipate ordering one "set" of computers. A set is a fairly balanced mixture of one Squid cache server, four or five Apache web servers and one database slave. We expect to add sets like this as needed, adjusting the exact balance as necessary.

The actual October order includes:

  • Two database/search servers because search and sometimes watchlists or history are operations which slow down contributors and to help to establish whether SCSI drives are worthwhile for database slaves. RAID 0 because we don't care whether a drive failure loses all of the data on a database slave. Many other computers have the same data.
  • Additional RAM to upgrade the first generation of Florida Apache web servers to 4GB of RAM will increase the amount of RAM for caching pages for logged in contributors and should reduce the Apache CPU and database load.
  • Two dual use Apache/DB backup slave machines with a total of 500GB of SATA disk space are ordered because the existing machines with 200GB of disk space don't have sufficient space for the job. The existing pair will be used as web servers only.
  • For Apaches on 12 October, with 12 total at about 1,000 total requests per second:
    • 11 active, CPU use was 90%-95% plus.
    • 12 active, reduced that to about 85-90%.
    • Adding browne from Squid service (the other five can handle load at present) reduced it further, to 65-80% (variation in Apache load increases at lower CPU levels).
    • The next day, 13 October, with about 1,100 requests per second and 13 active Apaches, 5 active Squids. Apache CPUs were about 65, 65, 65, 65, 70, 70, 75, 75, 80, 55, 65, 80, 80 (average 70%) and Squids 75, 75, 75, 80, 65 (average 74%). The site was fast.
    • From this order, the two DB/Apache machines, two to replace Apaches already moved to Squid service and one more should be sufficient for the next 30 days. Likely CPU load would be 50-60% or less at 1,000 requests per second, ignoring expected improvements in caching from the extra RAM.
    • Expect to order 3-5 more in each of November and December.
  • Tests on Squids have shown them to be limited by the speed of saving newly produced pages to their hard drives. Two 10,000 RPPM SATA drives will allow testing of two Squids with a pair of drives, one with 10,000RPM SATA and the other with 7200RPM SATA. If tests go well, we are likely to switch to Squids with many more drives and move some of the existing Squids back to Apache and memcached or tugelacache service.

Silicon Mechanics exact specs[edit]

Comments in italics.

1 SM-2280SATA "SATA 10K DB slave"
CPU:  Dual AMD Opteron 242 - 1.6 GHz - 1MB L2 Cache
RAM:  4GB (4 x 1GB) PC2700 Registered ECC
PCI 1:  3Ware 9500S-8 8 Port SATA RAID Controller
HDD 1:  Western Digital Raptor 74GB (10Krpm - 8MB Cache) SATA
HDD 2:  Western Digital Raptor 74GB (10Krpm - 8MB Cache) SATA
HDD 3:  Western Digital Raptor 74GB (10Krpm - 8MB Cache) SATA
HDD 4:  Western Digital Raptor 74GB (10Krpm - 8MB Cache) SATA
HDD 5:  Western Digital Raptor 74GB (10Krpm - 8MB Cache) SATA
HDD 6:  Western Digital Raptor 74GB (10Krpm - 8MB Cache) SATA
Low Profile CD:  Slimline 24X CD-ROM
Floppy:  1.44MB Floppy
Power Supply:  460W Hotswap Redundant
Rail Kit:  2U 3 Piece Sliding Rail Kit 24"-31"
WARRANTY: Standard 3 Year - Return to Depot
NOTES:
   Set up as RAID 0, stripe size 256k if available, else
   next lower size available. Write buffering on. Please
   add 9500S-8 battery if not standard. Fedora Core 2
   2.6.8-1.521smp preferred - but we'll upgrade if
   necesary.
                                                  $5521.00  $5521.00 
1 SM-2280S "SCSI 10K DB slave"
CPU:  Dual AMD Opteron 242 - 1.6 GHz - 1MB L2 Cache
Memory:  4GB (4 x 1GB) PC2700 Registered ECC
HDD 1:  Seagate Cheetah 10K.6 73GB U320 10KRPM SCA SCSI
HDD 2:  Seagate Cheetah 10K.6 73GB U320 10KRPM SCA SCSI
HDD 3:  Seagate Cheetah 10K.6 73GB U320 10KRPM SCA SCSI
HDD 4:  Seagate Cheetah 10K.6 73GB U320 10KRPM SCA SCSI
HDD 5:  Seagate Cheetah 10K.6 73GB U320 10KRPM SCA SCSI
HDD 6:  Seagate Cheetah 10K.6 73GB U320 10KRPM SCA SCSI
SCSI Controller:  LSI MegaRAID SCSI 320-1 - 1 Channel U320 RAID 64MB  CACHE w/ BBU
Low Profile CD-ROM:  Slimline 24X CD-ROM
Floppy:  1.44MB Floppy
Power Supply:  460W Hotswap Redundant
O/S:  Fedora Linux Core 1 - Preload, No Media
Rail Kit:  2U 3 Piece Sliding Rail Kit 24"-31"
WARRANTY: Standard 3 Year - Return to Depot
NOTES:
   Set up as RAID 0, stripe size 256k if available, else
   next lower size available. Write buffering on. Fedora
   Core 2 2.6.8-1.521smp preferred - but we'll upgrade if
   necesary.
                                                 $6316.00  $6316.00 

For the DB slaves, 4GB RAM and for SCSI 73GB drives because the price reduction saves almost enough for another SATA slave, which would deliver greater benefit. Don't really want to order that additional slave until tests show which drive option is best.

2 SM-1151SATA "Dual use Apache and backup-only DB slave"
CPU:  Intel P4 3.0GHz - 1MB Cache - HT - 800 FSB
RAM:  1GB (2 x 512MB) Unbuff ECC DDR 400 Interleaved
HDD 1:  Western Digital RE 250GB (7.2Krpm-8MB Cache) SATA
HDD 2:  Western Digital RE 250GB (7.2Krpm-8MB Cache) SATA
CD-ROM:  Slimline 24X CD-ROM
Floppy:  1.44MB Floppy
O/S:  Fedora Linux Core 2 - Preload, No Media
Rail Kit:  2 Piece Ball Bearing Rail Kit
WARRANTY: Standard 3 Year - Return to Depot
NOTES:
   Drives in RAID 0 please.
                                                 $1605.00  $3210.00 

These will replace the two currently in service in this role, saving jwales the setup time for adding a second drive in RAID 0 to the existing systems. 250GB drives to be at least as large as the other DB slaves. 1GB RAM to help the DB slave side keep up.

3 SM-1151SATA "Apache/ to replace some moved to Squid"
CPU:  Intel P4 3.0GHz - 1MB Cache - HT - 800 FSB
RAM:  1GB (2 x 512MB) Unbuff ECC DDR 400 Interleaved
HDD 1:  Seagate 200GB 7200.7 (7.2Krpm-8MB Cache) SATA
CD-ROM:  Slimline 24X CD-ROM
Floppy:  1.44MB Floppy
O/S:  Fedora Linux Core 2 - Preload, No Media
Rail Kit:  2 Piece Ball Bearing Rail Kit
WARRANTY: Standard 3 Year - Return to Depot
NOTES:
   None
                                                  $1254.00  $3762.00 

200GB drives instead of 250GB because the 250GB cost US$105 more while another 200GB drive in RAID 0 would cost $142 more. Quick poll of developers suggested no need for the extra 50GB or 200GB. So, save $315 for later use. switch from 2.8 to 3.0GHz CPU costs $6; worthwhile. 1GB of RAM to allow for tugelacache.

                                                  Total:   $18809.00 

In addition to this order from Silicon Mechanics the following purchases are planned:

  • Two 37GB Western Digital Raptor drives to go into one of the existing Squids, the existing drive to be moved to an identical squid and then comparative benchmarking to determine the effect on cache miss penalty determined. The smaller drive size because the high drive option for squids would have perhaps 12 drives and the idea is to get as many independently moving heads as possible, not capacity.
  • Additional RAM for the first generation web servers, to bring them all to 4GB so they are available for memcached or tugelacache and Squid duty.

Earlier draft[edit]

From the vendor we have been using, we'd order this:

  • 1x Squid:
    • Single P4 CPU around 2.4-3.0GHz (we don't care about AMD/Intel)
    • 4GB of RAM
    • Two 10,000 RPM SATA drives (we want to test performance of these in cache miss situations). We would try more if the case we normally used supported more.
  • SM-1151SATA P4 3.2GHz, 4x1GBMB ECC DDR400 RAM, Seagate 200GB drive, CD-ROM, floppy, $2462 This time we'll actually get this with 1GB and use it as a web server, replacing an older machine being upgraded to 4GB and moved to Squid service, using the two drives from this one.
  • 5x Web servers:
    • Single P4 CPU
    • 1GB of RAM (1GB this time)
    • 200GB or so SATA drive for log/utility storage.
  • SM-1151SATA P4 3.2GHz, 2x512MB ECC DDR400 RAM, Seagate 200GB drive, CD-ROM, floppy, $1361 each, $5444 total
  • 1x Database slave (to be configured for search):
    • Dual Opteron CPU, slower end of the range (we don't care about make, want to be sure we don't run out of CPU, though actual use is typically 20%)
    • 4GB of RAM (8 would be helpful)
    • 6 x 10,000 RPM SATA drives in RAID 0. We know that 7200RPM in RAID 1 does well, we want to see how this configuration does and we don't care about losing it all if a drive dies - it's just a slave, won't ever see data which isn't already safely stored elsewhere)
    • Write caching RAID controller (very important for database speed)
  • SM-2280SATA, Dual Opteron 242 1.6GHz, 4x1GB ECC DDR400 RAM, 6x Western Digital Raptor 10,000 RPM 74GB SATA drives in RAID 0, CD-ROM, floppy, $5352
  • Provisional: Cisco Catalyst 2948G-GE-TX [1] ($3,500).
  • TOTAL: $16,758