All new accounts created on Gitlab now require administrator approval. If you invite any collaborators, please let Flux staff know so they can approve the accounts.

hardware.scrbl 17 KB
Newer Older
Robert Ricci's avatar
Robert Ricci committed
1
#lang scribble/manual
2
@(require "defs.rkt")
Robert Ricci's avatar
Robert Ricci committed
3

4
@title[#:tag "hardware" #:version apt-version]{Hardware}
Robert Ricci's avatar
Robert Ricci committed
5

6
@(tb) can allocate experiments on any one of several federated clusters.
7 8

@clab-only{
Robert Ricci's avatar
Robert Ricci committed
9 10 11
    @(tb) has the ability to dispatch experiments to several clusters: three
    that belong to CloudLab itself, plus several more that belong to federated
    projects.
12

13 14
    Additional hardware expansions are planned, and descriptions of them
    can be found at @url[(@apturl "hardware.php")]
15
}
16

17
@clab-only{
Robert Ricci's avatar
Robert Ricci committed
18
    @; Reminder! This is duplicated below for Apt
19
    @section[#:tag "cloudlab-utah"]{CloudLab Utah}
20

Robert Ricci's avatar
Robert Ricci committed
21
    The CloudLab cluster at the University of Utah is being built in partnership
22
    with HP. It consists of 315 64-bit ARM servers and 270 Intel Xeon-D severs.
23 24 25
    Each has 8 cores, for a total of 4,680 cores. The servers are
    built on HP's Moonshot platform.
    The cluster is housed in the University of Utah's Downtown
26 27
    Data Center in Salt Lake City.

Robert Ricci's avatar
Robert Ricci committed
28
    More technical details can be found at @url[(@apturl "hardware.php#utah")]
29

30
    @(nodetype "m400" 315 "64-bit ARM"
31 32 33 34 35 36
        (list "CPU"  "Eight 64-bit ARMv8 (Atlas/A57) cores at 2.4 GHz (APM X-GENE)")
        (list "RAM"  "64GB ECC Memory (8x 8 GB DDR3-1600 SO-DIMMs)")
        (list "Disk" "120 GB of flash (SATA3 / M.2, Micron M500)")
        (list "NIC"  "Dual-port Mellanox ConnectX-3 10 GB NIC (PCIe v3.0, 8 lanes")
    )

37 38 39
    @(nodetype "m510" 270 "Intel Xeon-D"
        (list "CPU"  "Eight-core Intel Xeon D-1548 at 2.0 GHz")
        (list "RAM"  "64GB ECC Memory (4x 16 GB DDR4-2133 SO-DIMMs)")
Robert Ricci's avatar
Robert Ricci committed
40
        (list "Disk" "256 GB NVMe flash storage")
41 42 43
        (list "NIC"  "Dual-port Mellanox ConnectX-3 10 GB NIC (PCIe v3.0, 8 lanes")
    )

Robert Ricci's avatar
Robert Ricci committed
44
    There are 45 nodes in a chassis, and this cluster consists of thirteen
45 46 47 48 49 50 51 52 53 54 55
    chassis. Each chassis has two 45XGc switches; each node is connected to
    both switches, and each chassis switch has four 40Gbps uplinks, for a total
    of 320Gbps of uplink capacity from each chassis. One switch is used for
    control traffic, connecting to the Internet, etc. The other is used to
    build experiment topologies, and should be used for most experimental
    purposes.

    All chassis are interconnected through a large HP FlexFabric 12910 switch
    which has full bisection bandwidth internally.

    We have plans to enable some users to allocate entire chassis; when
Robert Ricci's avatar
Robert Ricci committed
56
    allocated in this mode, it will be possible to have complete administrator
57
    control over the switches in addition to the nodes.
58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80

    In phase two we added 50 Apollo R2200 chassis each with four HPE ProLiant
    XL170r server modules. Each server has 10 cores for a total of 2000 cores.

    @(nodetype "xl170" 200 "Intel Broadwell, 10 core, 1 disk"
        (list "CPU"  "Ten-core Intel E5-2640v4 at 2.4 GHz")
        (list "RAM"  "64GB ECC Memory (4x 16 GB DDR4-2400 DIMMs)")
        (list "Disk" "Intel DC S3520 480 GB 6G SATA SSD")
        (list "NIC"  "Two Dual-port Mellanox ConnectX-4 25 GB NIC (PCIe v3.0, 8 lanes")
    )

    Each server is connected via a 10Gbps control link and a 25Gbps experimental
    link to leaf Dell and Mellanox switches in groups of 40 servers.
    Each of the five groups' experimental switches are connected to a Mellanox
    2700 spine switch at 5x100Gbps. That switch in turn interconnects with the
    rest of the Utah CloudLab cluster via 6x40Gbps uplinks to the HP FlexFabric
    12910 switch.

    A unique feature of the phase two nodes is the addition of eight ONIE
    bootable "user allocatable" switches that can run a variety of Open Network
    OSes. These switches and all 200 nodes are connected to two NetScout 3903
    layer-1 switches, allowing flexible combinations of nodes and switches
    in an experiment.
81 82
}

Robert Ricci's avatar
Robert Ricci committed
83
@clab-only{
84
    @section[#:tag "cloudlab-wisconsin"]{CloudLab Wisconsin}
Robert Ricci's avatar
Robert Ricci committed
85

86 87
    The CloudLab cluster at the University of Wisconsin is built in
    partnership with Cisco, Seagate, and HP. The cluster, which is in
88 89 90
    Madison, Wisconsin, has 270 servers with a total of 5,000 cores connected
    in a CLOS topology with full bisection bandwidth. It has 1,070 TB of
    storage, including SSDs on every node. 
Robert Ricci's avatar
Robert Ricci committed
91

92 93
    @bold{Note: As of the time of writing, Wiscosin is working on upgrading nodes, and some may still be listed under their old ``type''}

Robert Ricci's avatar
Robert Ricci committed
94
    More technical details can be found at @url[(@apturl "hardware.php#wisconsin")]
Robert Ricci's avatar
Robert Ricci committed
95

96
    @(nodetype "c220g1" 90 "Haswell, 16 core, 3 disks"
Robert Ricci's avatar
Robert Ricci committed
97
        (list "CPU"  "Two Intel E5-2630 v3 8-core CPUs at 2.40 GHz (Haswell w/ EM64T)")
98 99
        (list "RAM"  "128GB ECC Memory (8x 16 GB DDR4 1866 MHz dual rank RDIMMs)")

Robert Ricci's avatar
Robert Ricci committed
100 101

        (list "Disk" "Two 1.2 TB 10K RPM 6G SAS SFF HDDs")
102 103
        (list "Disk" "One Intel DC S3500 480 GB 6G SATA SSDs")
        (list "NIC"  "Dual-port Intel X520-DA2 10Gb NIC (PCIe v3.0, 8 lanes)")
Robert Ricci's avatar
Robert Ricci committed
104 105
        (list "NIC"  "Onboard Intel i350 1Gb"))

106
    @(nodetype "c240g1" 10 "Haswell, 16 core, 14 disks"
Robert Ricci's avatar
Robert Ricci committed
107 108
        (list "CPU"  "Two Intel E5-2630 v3 8-core CPUs at 2.40 GHz (Haswell w/ EM64T)")

109 110 111
        (list "RAM"  "128GB ECC Memory (8x 16 GB DDR4 1866 MHz dual rank RDIMMs)")

        (list "Disk" "Two Intel DC S3500 480 GB 6G SATA SSDs")
Robert Ricci's avatar
Robert Ricci committed
112
        (list "Disk" "Twelve 3 TB 3.5\" HDDs donated by Seagate")
113
        (list "NIC"  "Dual-port Intel X520-DA2 10Gb NIC (PCIe v3.0, 8 lanes)")
Robert Ricci's avatar
Robert Ricci committed
114 115
        (list "NIC"  "Onboard Intel i350 1Gb"))

116
    @(nodetype "c220g2" 163 "Haswell, 20 core, 3 disks"
117
        (list "CPU"  "Two Intel E5-2660 v3 10-core CPUs at 2.60 GHz (Haswell EP)")
118
        (list "RAM"  "160GB ECC Memory (10x 16 GB DDR4 2133 MHz dual rank RDIMMs)")
119
        (list "Disk" "One Intel DC S3500 480 GB 6G SATA SSDs")
120 121 122 123
        (list "Disk" "Two 1.2 TB 10K RPM 6G SAS SFF HDDs")
        (list "NIC"  "Dual-port Intel X520 10Gb NIC (PCIe v3.0, 8 lanes")
        (list "NIC"  "Onboard Intel i350 1Gb"))

124
    @(nodetype "c240g2" 7 "Haswell, 20 core, 14 disks"
125
        (list "CPU"  "Two Intel E5-2660 v3 10-core CPUs at 2.60 GHz (Haswell EP)")
126
        (list "RAM"  "160GB ECC Memory (10x 16 GB DDR4 2133 MHz dual rank RDIMMs)")
127
        (list "Disk" "Two Intel DC S3500 480 GB 6G SATA SSDs")
128 129 130 131
        (list "Disk" "Twelve 3 TB 3.5\" HDDs donated by Seagate")
        (list "NIC"  "Dual-port Intel X520 10Gb NIC (PCIe v3.0, 8 lanes")
        (list "NIC"  "Onboard Intel i350 1Gb"))

Robert Ricci's avatar
Robert Ricci committed
132 133
    All nodes are connected to two networks:

134 135 136
    @margin-note{The experiment network at Wisconsin is transitioning to using
        HP switches in order to provide OpenFlow 1.3 support.}

Robert Ricci's avatar
Robert Ricci committed
137 138 139 140 141 142 143 144 145 146
    @itemlist[
        @item{A 1 Gbps Ethernet @bold{``control network''}---this network 
            is used for remote access, experiment management, etc., and is
            connected to the public Internet. When you log in to nodes in your
            experiment using @code{ssh}, this is the network you are using.
            @italic{You should not use this network as part of the experiments
            you run in @(tb).}
        }

        @item{A 10 Gbps Ethernet @bold{``experiment network''}--each node has
Robert Ricci's avatar
Robert Ricci committed
147
            @bold{two interfaces} on this network. Twelve leaf switches are
Robert Ricci's avatar
Robert Ricci committed
148 149 150 151 152 153 154 155 156 157
            Cisco Nexus C3172PQs, which have 48 10Gbps ports for the nodes and 
            six 40Gbps uplink ports. They are connected to six spine switches 
            (Cisco Nexus C3132Qs); each leaf has one 40Gbps link to each
            spine switch. Another C3132Q switch acts as a core; each spine
            switch has one 40Gbps link to it, and it has upstream
            links to Internet2.
        }
    ]
}

158
@clab-only{
159
    @section[#:tag "cloudlab-clemson"]{CloudLab Clemson}
160

161 162 163
    The CloudLab cluster at Clemson University has been built
    partnership with Dell. The cluster so far has 186
    servers with a total of 4,400 cores, 596TB of disk space, and
164 165
    48TB of RAM. All nodes have 10GB Ethernet and QDR
    Infiniband.  It is located in Clemson, South Carolina.
166 167 168

    More technical details can be found at @url[(@apturl "hardware.php#clemson")]

169
    @(nodetype "c8220" 96 "Ivy Bridge, 20 core"
170 171 172 173 174 175 176
        (list "CPU"  "Two Intel E5-2660 v2 10-core CPUs at 2.20 GHz (Ivy Bridge)")
        (list "RAM"  "256GB ECC Memory (16x 16 GB DDR4 1600MT/s dual rank RDIMMs")

        (list "Disk" "Two 1 TB 7.2K RPM 3G SATA HDDs")
        (list "NIC"  "Dual-port Intel 10Gbe NIC (PCIe v3.0, 8 lanes")
        (list "NIC"  "Qlogic QLE 7340 40 Gb/s Infiniband HCA (PCIe v3.0, 8 lanes)"))

177
    @(nodetype "c8220x" 4 "Ivy Bridge, 20 core, 20 disks"
178 179 180 181
        (list "CPU"  "Two Intel E5-2660 v2 10-core CPUs at 2.20 GHz (Ivy Bridge)")
        (list "RAM"  "256GB ECC Memory (16x 16 GB DDR4 1600MT/s dual rank RDIMMs")

        (list "Disk" "Eight 1 TB 7.2K RPM 3G SATA HDDs")
182
        (list "Disk" "Twelve 4 TB 7.2K RPM 3G SATA HDDs")
183 184 185
        (list "NIC"  "Dual-port Intel 10Gbe NIC (PCIe v3.0, 8 lanes")
        (list "NIC"  "Qlogic QLE 7340 40 Gb/s Infiniband HCA (PCIe v3.0, 8 lanes)"))

186
    @(nodetype "c6320" 84 "Haswell, 28 core"
187 188 189
        (list "CPU"  "Two Intel E5-2683 v3 14-core CPUs at 2.00 GHz (Haswell)")
        (list "RAM"  "256GB ECC Memory")
        (list "Disk" "Two 1 TB 7.2K RPM 3G SATA HDDs")
Robert Ricci's avatar
Robert Ricci committed
190
        (list "NIC"  "Dual-port Intel 10Gbe NIC (X520)")
191
        (list "NIC"  "Qlogic QLE 7340 40 Gb/s Infiniband HCA (PCIe v3.0, 8 lanes)"))
192

193
    @(nodetype "c4130" 2 "Haswell, 28 core, two GPUs"
194 195 196 197 198
        (list  "CPU"  "Two Intel E5-2680 v3 12-core processors at 2.50 GHz (Haswell)")
        (list "RAM"  "256GB ECC Memory")
        (list "Disk" "Two 1 TB 7.2K RPM 3G SATA HDDs")
        (list "GPU"  "Two Tesla K40m GPUs")
        (list "NIC"  "Dual-port Intel 1Gbe NIC (i350)")
Robert Ricci's avatar
Robert Ricci committed
199
        (list "NIC"  "Dual-port Intel 10Gbe NIC (X710)")
200
        (list "NIC"  "Qlogic QLE 7340 40 Gb/s Infiniband HCA (PCIe v3.0, 8 lanes)"))
201

Mike Hibler's avatar
Mike Hibler committed
202 203 204 205 206 207 208 209 210 211 212
    There are also two, storage intensive (270TB each!) nodes
    that should only be used if you need a huge amount of volatile
    storage. These nodes have only 10GB Ethernet.

    @(nodetype "dss7500" 2 "Haswell, 12 core, 270TB disk"
        (list "CPU"  "Two Intel E5-2620 v3 6-core CPUs at 2.40 GHz (Haswell)")
        (list "RAM"  "128GB ECC Memory")
        (list "Disk" "Two 120 GB 6Gbps SATA SSDs")
        (list "Disk" "45 6 TB 7.2K RPM 6Gbps SATA HDDs")
        (list "NIC"  "Dual-port Intel 10Gbe NIC (X520)"))

213
    There are three networks at the Clemson site:
214 215 216 217 218 219 220 221 222 223 224

    @itemlist[
        @item{A 1 Gbps Ethernet @bold{``control network''}---this network 
            is used for remote access, experiment management, etc., and is
            connected to the public Internet. When you log in to nodes in your
            experiment using @code{ssh}, this is the network you are using.
            @italic{You should not use this network as part of the experiments
            you run in @(tb).}
        }

        @item{A 10 Gbps Ethernet @bold{``experiment network''}--each node has
225 226 227 228
            @bold{one interface} on this network.  This network is implemented
            using three Force10 S6000 and three Force10 Z9100 switches.  Each
            S6000 switch is connected to a companion Z9100 switch via a 480Gbps
            link aggregate.
229 230
        }

231 232
        @item{A 40 Gbps QDR Infiniband @bold{``experiment network''}--each
            node has one connection to this
233 234 235
            network, which is implemented using a large Mellanox chassis switch
            with full bisection bandwidth.
        }
236 237 238
    ]
}

239 240
@section[#:tag "apt-cluster"]{Apt Cluster}

241 242 243
@apt-only{@margin-note{This is the cluster that is currently used by default
    for all experiments on Apt.}}

Robert Ricci's avatar
Robert Ricci committed
244 245 246
@clab-only{@margin-note{This cluster is not owned by CloudLab, but is federated
and available to CloudLab users.}}

247 248 249
The main Apt cluster is housed in the University of Utah's Downtown Data
Center in Salt Lake City, Utah. It contains two classes of nodes:

250
@(nodetype "r320" 128 "Sandy Bridge, 8 cores"
251 252 253 254 255 256 257
    (list "CPU"   "1x Xeon E5-2450 processor (8 cores, 2.1Ghz)")
    (list "RAM"   "16GB Memory (4 x 2GB RDIMMs, 1.6Ghz)")
    (list "Disks" "4 x 500GB 7.2K SATA Drives (RAID5)")
    (list "NIC"   "1GbE Dual port embedded NIC (Broadcom)")
    (list "NIC"   "1 x Mellanox MX354A Dual port FDR CX3 adapter w/1 x QSA adapter")
)

258
@(nodetype "c6220" 64 "Ivy Bridge, 16 cores"
259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302
    (list "CPU"   "2 x Xeon E5-2650v2 processors (8 cores each, 2.6Ghz)")
    (list "RAM"   "64GB Memory (8 x 8GB DDR-3 RDIMMs, 1.86Ghz)")
    (list "Disks" "2 x 1TB SATA 3.5” 7.2K rpm hard drives")
    (list "NIC"   "4 x 1GbE embedded Ethernet Ports (Broadcom)")
    (list "NIC"   "1 x Intel X520 PCIe Dual port 10Gb Ethernet NIC")
    (list "NIC"   "1 x Mellanox FDR CX3 Single port mezz card")
)

All nodes are connected to three networks with @bold{one interface each}:

@itemlist[
    @item{A 1 Gbps @italic{Ethernet} @bold{``control network''}---this network
        is used for remote access, experiment management, etc., and is
        connected to the public Internet. When you log in to nodes in your
        experiment using @code{ssh}, this is the network you are using.
        @italic{You should not use this network as part of the experiments you
        run in Apt.}
    }

    @item{A @bold{``flexible fabric''} that can run up to 56 Gbps and runs
        @italic{either FDR Infiniband or Ethernet}. This fabric uses NICs and
        switches with @hyperlink["http://www.mellanox.com/"]{Mellanox's}
        VPI technology. This means that we can, on demand, configure each
        port to be either FDR Inifiniband or 40 Gbps (or even non-standard
        56 Gbps) Ethernet. This fabric consists of seven edge switches
        (Mellanox SX6036G) with 28 connected nodes each. There are two core
        switches (also SX6036G), and each edge switch connects to both cores
        with a 3.5:1 blocking factor. This fabric is ideal if you need
        @bold{very low latency, Infiniband, or a few, high-bandwidth Ethernet
        links}.
     }

    @item{A 10 Gbps @italic{Ethernet} @bold{``commodity fabric''}.  One the
        @code{r320} nodes, a port on the Mellanox NIC (permanently set to
        Ethernet mode) is used to connect to this fabric; on the @code{c6220}
        nodes, a dedicated Intel 10 Gbps NIC is used. This fabric is built 
        from two Dell Z9000 switches, each of which has 96 nodes connected
        to it. It is idea for creating @bold{large LANs}: each of the two
        switches has full bisection bandwidth for its 96 ports, and there is a
        3.5:1 blocking factor between the two switches.
    }
        
]

Robert Ricci's avatar
Robert Ricci committed
303
@section[#:tag "ig-ddc"]{IG-DDC Cluster}
304

Robert Ricci's avatar
Robert Ricci committed
305 306 307
@clab-only{@margin-note{This cluster is not owned by CloudLab, but is federated
and available to CloudLab users.}}

308
This small cluster is an @hyperlink["http://www.instageni.net"]{InstaGENI
309
Rack} housed in the University of Utah's Downtown Data Center. It
310 311
has nodes of only a single type:

312
@nodetype["dl360" 33 "Sandy Bridge, 16 cores"
313 314 315 316 317
    (list "CPU"   "2x Xeon E5-2450 processors (8 cores each, 2.1Ghz)")
    (list "RAM"   "48GB Memory (6 x 8GB RDIMMs, 1.6Ghz)")
    (list "Disk"  "1 x 1TB 7.2K SATA Drive")
    (list "NIC"   "1GbE 4-port embedded NIC")
]
318 319 320 321

It has two network fabrics:

@itemlist[
322 323
    @item{A 1 Gbps @bold{``control network''}. 
        This is used for remote access,
Robert Ricci's avatar
Robert Ricci committed
324
        and should not be used for experiments.}
325 326 327 328
    @item{A 1 Gbps @bold{``experiment network''}. Each node has @italic{three}
        interfaces on this network, which is built from a single HP Procurve
        5406 switch. OpenFlow is available on this network.}
]
Robert Ricci's avatar
Robert Ricci committed
329 330 331

@; Total hack, I don't want this cluster listed first for Apt
@apt-only{
Robert Ricci's avatar
Robert Ricci committed
332
    @section[#:tag "cloudlab-utah"]{CloudLab Utah}
Robert Ricci's avatar
Robert Ricci committed
333 334 335 336 337 338 339 340 341 342 343 344 345 346

    @margin-note{This cluster is part of
    @link["https://www.cloudlab.us"]{CloudLab}, but is also available to Apt
    users.}

    The CloudLab cluster at the University of Utah is being built in partnership
    with HP. The first phase of this cluster consists of 315 64-bit ARM
    servers with 8 cores each, for a total of 2,520 cores. The servers are
    built on HP's Moonshot platform using X-GENE system-on-chip designs from
    Applied Micro. The cluster is hosted in the University of Utah's Downtown
    Data Center in Salt Lake City.

    More technical details can be found at @url[(@apturl "hardware.php#utah")]

347
    @(nodetype "m400" 315 "64-bit ARM"
Robert Ricci's avatar
Robert Ricci committed
348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368
        (list "CPU"  "Eight 64-bit ARMv8 (Atlas/A57) cores at 2.4 GHz (APM X-GENE)")
        (list "RAM"  "64GB ECC Memory (8x 8 GB DDR3-1600 SO-DIMMs)")
        (list "Disk" "120 GB of flash (SATA3 / M.2, Micron M500)")
        (list "NIC"  "Dual-port Mellanox ConnectX-3 10 GB NIC (PCIe v3.0, 8 lanes")
    )

    There are 45 nodes in a chassis, and this cluster consists of seven
    chassis. Each chassis has two 45XGc switches; each node is connected to
    both switches, and each chassis switch has four 40Gbps uplinks, for a total
    of 320Gbps of uplink capacity from each chassis. One switch is used for
    control traffic, connecting to the Internet, etc. The other is used to
    build experiment topologies, and should be used for most experimental
    purposes.

    All chassis are interconnected through a large HP FlexFabric 12910 switch
    which has full bisection bandwidth internally.

    We have plans to enable some users to allocate entire chassis; when
    allocated in this mode, it will be possible to have complete administrator
    control over the switches in addition to the nodes.
}