1 |
On 05/19/2016 06:47 PM, Robin H. Johnson wrote: |
2 |
> Infra has already discussed most of this hardware planning in |
3 |
> #gentoo-infra, but I thought it might be useful to see any other |
4 |
> comments on the hardware plan. If you wish to make private comments to |
5 |
> this thread, please send directly to infra@g.o or |
6 |
> gentoo-core@l.g.o instead of the gentoo-project list. |
7 |
> |
8 |
> Remarks like 'you should use ZFS instead of this' aren't directly |
9 |
> helpful to this discussion. What is more useful is pointing out any |
10 |
> potential problems you might see with the plan, or gotchas in the |
11 |
> hardware. |
12 |
> |
13 |
> We've previously run Ganeti [0] with general success, and we'd like to |
14 |
> continue doing so (vs libvirt or openstack). It offers VM storage |
15 |
> redundancy via DRBD (amongst other options), which we're going to take |
16 |
> best advantage of by using a cross-over 10Gbit link between two nodes |
17 |
> (as we have no 10GBit switching in the environment). Some of the VMs |
18 |
> will run on spinning disk, others on SSD, others maybe w/ dm-cache. |
19 |
> libvirt IS an easy fallback from Ganeti, but lacks some of the automated |
20 |
> failover and DRBD handling options. |
21 |
> |
22 |
> This will house at least the following existing VMs, all of which have |
23 |
> large storage needs: |
24 |
> - woodpecker.gentoo.org |
25 |
> - roverlay.dev.g.o |
26 |
> - tinderbox.amd64.dev.g.o |
27 |
> - devbox.amd64.dev.g.o |
28 |
> |
29 |
> And virtualize the following older systems: |
30 |
> [2007 Dells] |
31 |
> - finch.g.o (puppet) |
32 |
> - vulture.g.o (GSoC host) |
33 |
> [2010 Atoms] |
34 |
> - bellbird.g.o (infra services) |
35 |
> - bittern.g.o (blogs webhost) |
36 |
> - bobolink.g.o (rsync.g.o node, dns slave) |
37 |
> - brambling.g.o (bouncer, devmanual, infra-status) |
38 |
> [Other] |
39 |
> - meadowlark.g.o (infra services) |
40 |
> |
41 |
> And New VMs/services: |
42 |
> - split git to rsync & snapshot generation from dipper? |
43 |
> - split blogs (and other) database hosting from dipper? |
44 |
> |
45 |
> We'd probably keep the two other 2011 Dell systems in operation for the |
46 |
> moment, to distribute load better, but have enough capacity to run their |
47 |
> VMs as when they fail. |
48 |
> |
49 |
> The general best prices we've seen are from a vendor that's new to us, |
50 |
> WiredZone, and we're willing to give them a try unless somebody has even |
51 |
> better pricing to offer us. |
52 |
> |
53 |
> Hardware (all in $USD): |
54 |
> Supermicro SYS-2028TP-DECTR [1][2] |
55 |
> - $2,732.42/ea, quantity 1 |
56 |
> - two half-width 2U nodes in a single chassis w/ shared redundant PSU. |
57 |
> - each node has: |
58 |
> - 2x 10GBe ports (there are no SFP options) |
59 |
> - 12x 2.5" SAS3, controller in JBOD/IT mode |
60 |
> Per node: |
61 |
> Intel Xeon E5-2620v4 [3] - |
62 |
> - $421.56/ea, quantity 2 |
63 |
> 32GB DDR4 PC4-19200 (2400MHz) 288-pin RDIMM ECC Registered [4], |
64 |
> - $162.89/ea, quantity 4 |
65 |
> - require min of two DIMMs per CPU |
66 |
> - price jump to 64GB DIMMs very high. |
67 |
> - buy more RAM later? |
68 |
> Seagate 2TB SAS 12Gb/s 7200RPM 2.5in, ST2000NX0273 [5] |
69 |
> - $315.18/ea, quantity 4 |
70 |
> - 4-disk RAID5 (mdadm) |
71 |
> Samsung 850 EVO 1TB, MZ-75E1T0B/AM [6] |
72 |
> - $345.00/ea, quantity 2 |
73 |
> - RAID1 (mdadm) |
74 |
> = $3445.40/node |
75 |
> |
76 |
> Overall cost: |
77 |
> $2,732.42 - chassis |
78 |
> $3,445.40 - left node components |
79 |
> $3,445.40 - right node components |
80 |
> $ 315.18 - 1x spare ST2000NX0273 HDD |
81 |
> $ 25.00 - 3ft CAT6a patch cable (estimated) |
82 |
> |
83 |
> Parts sub-total: $9,963.40 |
84 |
> Labour sub-total: $300 (estimate) |
85 |
> Taxes: $0.00 (Oregon has no sales taxes) |
86 |
> S&H: $200 (estimate) |
87 |
> |
88 |
> Grant total: $10,463.40 (USD) |
89 |
> |
90 |
> Future hardware improvement options: |
91 |
> - Add more RAM |
92 |
> - Add up to 6x more disks per node. |
93 |
> |
94 |
> [0] http://www.ganeti.org/ |
95 |
> [1] http://www.supermicro.com/products/system/2U/2028/SYS-2028TP-DECTR.cfm |
96 |
> [2] http://www.wiredzone.com/supermicro-multi-node-servers-twin-barebone-dual-cpu-2-node-sys-2028tp-dectr-10024389 |
97 |
> [3] https://www.wiredzone.com/intel-components-cpu-processors-server-bx80660e52620v4-10025960 |
98 |
> [4] https://www.wiredzone.com/supermicro-components-memory-ddr4-mem-dr432l-sl01-er24-10025993 |
99 |
> [5] https://www.wiredzone.com/seagate-components-hard-drives-enterprise-st2000nx0273-10024175 |
100 |
> [6] https://www.wiredzone.com/samsung-components-hard-drives-enterprise-mz-75e1t0b-am-10024043 |
101 |
> |
102 |
|
103 |
+1 to this generally, the one question I have is if we want to spend ~1k |
104 |
more on one of the 4x nodes, it'd allow us to expand easier in the |
105 |
future. The reasoning against this that I can think of is that we want |
106 |
a higher disk/node ratio (which is just 6 per node instead of 12). |
107 |
|
108 |
-- |
109 |
-- Matthew Thode (prometheanfire) |