[Make-wifi-fast] Flent test hardware

[Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 2038 bytes --]

My Flent test rig needs a refresh. I currently use two Mac Minis. The Intel with the Core 2 Duo P7550 and forcedeth Ethernet is OK, so I could possibly keep it (but no BQL support which is not ideal for some tests), but it’s time for my G4 Mini with 100 Mbit Ethernet and stratospheric clock drift to go.

So I’m searching for one or two low-cost Flent devices (client or server- which needs more CPU by the way?). Requirements:

- Gigabit Ethernet (1x ok, 2x better) with a reliable Linux driver with BQL support
- PTP timestamp support (http://linuxptp.sourceforge.net <http://linuxptp.sourceforge.net/>), hardware preferable
- enough CPU to accurately do Flent’s higher flow count tests like rrul_torrent or rrul_be_nflows with 64 flows (sometimes I also do these together with one or more instances of rrul_be to test host fairness)

I started a spreadsheet of what I’ve found so far here in Czech:

https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit?usp=sharing <https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit?usp=sharing>

I’m leaning towards either one of the low-end Intel NUCs or GIGABYTE BRIX mini PCs, to which I’d add some memory and storage. But so many of these low-end devices come with Realtek Ethernet (r8169 driver?). I’m not sure how stable that driver is, if the BQL support is usable (https://www.bufferbloat.net/projects/bloat/wiki/BQL_enabled_drivers/ <https://www.bufferbloat.net/projects/bloat/wiki/BQL_enabled_drivers/>), and I’d have to settle for software timestamp support for PTP. With a higher budget, I might go for a 1U server with well-supported Intel NICs, something like described in this setup (http://www.academia.edu/10312557/DEVELOPING_LOW-COST_NTP_STRATUM_1_SERVERS_WITH_LINUX_PTP_AND_GPS <http://www.academia.edu/10312557/DEVELOPING_LOW-COST_NTP_STRATUM_1_SERVERS_WITH_LINUX_PTP_AND_GPS>). But I can’t seem to find those cheaply (even <$1000) here.

Any thoughts or ideas on this?


[-- Attachment #2: Type: text/html, Size: 2718 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Sebastian Moeller]

Hi Pete,

I like https://www.amazon.de/HP-ProLiant-MicroServer-G1610T-SATA-Server/dp/B013UBCHVU/ref=sr_1_1?ie=UTF8&qid=1509807007&sr=8-1&keywords=hp+proliant+microserver+gen8

quite a lot. It comes with 2 ethernet controllers and ports
03:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM5720 Gigabit Ethernet PCIe
03:00.1 Ethernet controller: Broadcom Corporation NetXtreme BCM5720 Gigabit Ethernet PCIe
as far as I can tell these support BQL, but I am not 100% sure...
and will also allow to replace the slightly anemic celeron with a xeon or an i3 or i5, at around 211EUR still a good catch, but a bit on the high side of your list, but it comes with 4GB ecc memory already...

Best Regards



> On Nov 4, 2017, at 14:33, Pete Heist <peteheist@gmail.com> wrote:
> 
> My Flent test rig needs a refresh. I currently use two Mac Minis. The Intel with the Core 2 Duo P7550 and forcedeth Ethernet is OK, so I could possibly keep it (but no BQL support which is not ideal for some tests), but it’s time for my G4 Mini with 100 Mbit Ethernet and stratospheric clock drift to go.
> 
> So I’m searching for one or two low-cost Flent devices (client or server- which needs more CPU by the way?). Requirements:
> 
> - Gigabit Ethernet (1x ok, 2x better) with a reliable Linux driver with BQL support
> - PTP timestamp support (http://linuxptp.sourceforge.net), hardware preferable
> - enough CPU to accurately do Flent’s higher flow count tests like rrul_torrent or rrul_be_nflows with 64 flows (sometimes I also do these together with one or more instances of rrul_be to test host fairness)
> 
> I started a spreadsheet of what I’ve found so far here in Czech:
> 
> https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit?usp=sharing
> 
> I’m leaning towards either one of the low-end Intel NUCs or GIGABYTE BRIX mini PCs, to which I’d add some memory and storage. But so many of these low-end devices come with Realtek Ethernet (r8169 driver?). I’m not sure how stable that driver is, if the BQL support is usable (https://www.bufferbloat.net/projects/bloat/wiki/BQL_enabled_drivers/), and I’d have to settle for software timestamp support for PTP. With a higher budget, I might go for a 1U server with well-supported Intel NICs, something like described in this setup (http://www.academia.edu/10312557/DEVELOPING_LOW-COST_NTP_STRATUM_1_SERVERS_WITH_LINUX_PTP_AND_GPS). But I can’t seem to find those cheaply (even <$1000) here.
> 
> Any thoughts or ideas on this?
> 
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 1602 bytes --]


> On Nov 4, 2017, at 4:06 PM, Sebastian Moeller <moeller0@gmx.de> wrote:
> 
> Hi Pete,
> 
> I like https://www.amazon.de/HP-ProLiant-MicroServer-G1610T-SATA-Server/dp/B013UBCHVU/ref=sr_1_1?ie=UTF8&qid=1509807007&sr=8-1&keywords=hp+proliant+microserver+gen8
> 
> quite a lot. It comes with 2 ethernet controllers and ports
> 03:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM5720 Gigabit Ethernet PCIe
> 03:00.1 Ethernet controller: Broadcom Corporation NetXtreme BCM5720 Gigabit Ethernet PCIe
> as far as I can tell these support BQL, but I am not 100% sure...
> and will also allow to replace the slightly anemic celeron with a xeon or an i3 or i5, at around 211EUR still a good catch, but a bit on the high side of your list, but it comes with 4GB ecc memory already...


That looks like a great option. It’s a little more than I’d hoped to spend, but also more than I expected to get for this price.

Can I ask, can you tell if it's using the tg3 driver, either with lsmod, or "dmesg | grep -i ethernet”, or “ethtool -i eth0”, or similar? If so, I see BQL support in that code, and it’s listed as having hardware timestamp support for PTP (http://linuxptp.sourceforge.net). Looks great...

I see a Gen 10 available here for slightly more with the same BCM5720 NIC, and a AMD Opteron X3216 instead, plus 8 gigs of RAM, which I don’t need really, but I can search more for similar hardware: https://www.alza.cz/hpe-proliant-microserver-gen10-d5127723.htm?o=1 <https://www.alza.cz/hpe-proliant-microserver-gen10-d5127723.htm?o=1>

Thanks for the tip!


[-- Attachment #2: Type: text/html, Size: 2609 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Sebastian Moeller]

Hi Pete,


> On Nov 4, 2017, at 17:33, Pete Heist <peteheist@gmail.com> wrote:
> 
> 
>> On Nov 4, 2017, at 4:06 PM, Sebastian Moeller <moeller0@gmx.de> wrote:
>> 
>> Hi Pete,
>> 
>> I like https://www.amazon.de/HP-ProLiant-MicroServer-G1610T-SATA-Server/dp/B013UBCHVU/ref=sr_1_1?ie=UTF8&qid=1509807007&sr=8-1&keywords=hp+proliant+microserver+gen8
>> 
>> quite a lot. It comes with 2 ethernet controllers and ports
>> 03:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM5720 Gigabit Ethernet PCIe
>> 03:00.1 Ethernet controller: Broadcom Corporation NetXtreme BCM5720 Gigabit Ethernet PCIe
>> as far as I can tell these support BQL, but I am not 100% sure...
>> and will also allow to replace the slightly anemic celeron with a xeon or an i3 or i5, at around 211EUR still a good catch, but a bit on the high side of your list, but it comes with 4GB ecc memory already...
> 
> That looks like a great option. It’s a little more than I’d hoped to spend, but also more than I expected to get for this price.
> 
> Can I ask, can you tell if it's using the tg3 driver,

Sure,



> either with lsmod, or "dmesg | grep -i ethernet”,

user@work-horse:~$ dmesg | grep -i ethernet
[    0.867370] tg3 0000:03:00.0 eth0: attached PHY is 5720C (10/100/1000Base-T Ethernet) (WireSpeed[1], EEE[1])
[    1.495403] tg3 0000:03:00.1 eth1: attached PHY is 5720C (10/100/1000Base-T Ethernet) (WireSpeed[1], EEE[1])


> or “ethtool -i eth0”,

user@work-horse:~$ ethtool -i eth0
driver: tg3
version: 3.137
firmware-version: 5720-v1.37 NCSI v1.3.12.0
expansion-rom-version: 
bus-info: 0000:03:00.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: no

user@work-horse:~$ ethtool -k eth0
Features for eth0:
rx-checksumming: on
tx-checksumming: on
	tx-checksum-ipv4: on
	tx-checksum-ip-generic: off [fixed]
	tx-checksum-ipv6: on
	tx-checksum-fcoe-crc: off [fixed]
	tx-checksum-sctp: off [fixed]
scatter-gather: on
	tx-scatter-gather: on
	tx-scatter-gather-fraglist: off [fixed]
tcp-segmentation-offload: on
	tx-tcp-segmentation: on
	tx-tcp-ecn-segmentation: on
	tx-tcp-mangleid-segmentation: off
	tx-tcp6-segmentation: on
udp-fragmentation-offload: off [fixed]
generic-segmentation-offload: on
generic-receive-offload: on
large-receive-offload: off [fixed]
rx-vlan-offload: on [fixed]
tx-vlan-offload: on [fixed]
ntuple-filters: off [fixed]
receive-hashing: off [fixed]
highdma: on
rx-vlan-filter: off [fixed]
vlan-challenged: off [fixed]
tx-lockless: off [fixed]
netns-local: off [fixed]
tx-gso-robust: off [fixed]
tx-fcoe-segmentation: off [fixed]
tx-gre-segmentation: off [fixed]
tx-gre-csum-segmentation: off [fixed]
tx-ipxip4-segmentation: off [fixed]
tx-ipxip6-segmentation: off [fixed]
tx-udp_tnl-segmentation: off [fixed]
tx-udp_tnl-csum-segmentation: off [fixed]
tx-gso-partial: off [fixed]
tx-sctp-segmentation: off [fixed]
fcoe-mtu: off [fixed]
tx-nocache-copy: off
loopback: off [fixed]
rx-fcs: off [fixed]
rx-all: off [fixed]
tx-vlan-stag-hw-insert: off [fixed]
rx-vlan-stag-hw-parse: off [fixed]
rx-vlan-stag-filter: off [fixed]
l2-fwd-offload: off [fixed]
busy-poll: off [fixed]
hw-tc-offload: off [fixed]



> or similar? If so, I see BQL support in that code,

	I believe sometime in the past I confirmed BQL working, but my memory is hazy and this might have been a different machine...

> and it’s listed as having hardware timestamp support for PTP (http://linuxptp.sourceforge.net). Looks great...
> 
> I see a Gen 10 available here for slightly more with the same BCM5720 NIC, and a AMD Opteron X3216 instead, plus 8 gigs of RAM, which I don’t need really, but I can search more for similar hardware: https://www.alza.cz/hpe-proliant-microserver-gen10-d5127723.htm?o=1

	Then gen10 microserver while newer comes with a soldered in pre-ZEN amd cpu and seems less extensible then the gen8, but then the gen10 probly is also a decent machine.


best regards
	Sebastian

> 
> Thanks for the tip!
> 

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 4258 bytes --]


> On Nov 4, 2017, at 6:25 PM, Sebastian Moeller <moeller0@gmx.de> wrote:
> 
> Hi Pete,
> 
>> or “ethtool -i eth0”,
> 
> user@work-horse:~$ ethtool -i eth0
> driver: tg3
> version: 3.137
> firmware-version: 5720-v1.37 NCSI v1.3.12.0
> expansion-rom-version: 
> bus-info: 0000:03:00.0
> supports-statistics: yes
> supports-test: yes
> supports-eeprom-access: yes
> supports-register-dump: yes
> supports-priv-flags: no
> 
> user@work-horse:~$ ethtool -k eth0
> Features for eth0:
> rx-checksumming: on
> tx-checksumming: on
> 	tx-checksum-ipv4: on
> 	tx-checksum-ip-generic: off [fixed]
> 	tx-checksum-ipv6: on
> 	tx-checksum-fcoe-crc: off [fixed]
> 	tx-checksum-sctp: off [fixed]
> scatter-gather: on
> 	tx-scatter-gather: on
> 	tx-scatter-gather-fraglist: off [fixed]
> tcp-segmentation-offload: on
> 	tx-tcp-segmentation: on
> 	tx-tcp-ecn-segmentation: on
> 	tx-tcp-mangleid-segmentation: off
> 	tx-tcp6-segmentation: on
> udp-fragmentation-offload: off [fixed]
> generic-segmentation-offload: on
> generic-receive-offload: on
> large-receive-offload: off [fixed]
> rx-vlan-offload: on [fixed]
> tx-vlan-offload: on [fixed]
> ntuple-filters: off [fixed]
> receive-hashing: off [fixed]
> highdma: on
> rx-vlan-filter: off [fixed]
> vlan-challenged: off [fixed]
> tx-lockless: off [fixed]
> netns-local: off [fixed]
> tx-gso-robust: off [fixed]
> tx-fcoe-segmentation: off [fixed]
> tx-gre-segmentation: off [fixed]
> tx-gre-csum-segmentation: off [fixed]
> tx-ipxip4-segmentation: off [fixed]
> tx-ipxip6-segmentation: off [fixed]
> tx-udp_tnl-segmentation: off [fixed]
> tx-udp_tnl-csum-segmentation: off [fixed]
> tx-gso-partial: off [fixed]
> tx-sctp-segmentation: off [fixed]
> fcoe-mtu: off [fixed]
> tx-nocache-copy: off
> loopback: off [fixed]
> rx-fcs: off [fixed]
> rx-all: off [fixed]
> tx-vlan-stag-hw-insert: off [fixed]
> rx-vlan-stag-hw-parse: off [fixed]
> rx-vlan-stag-filter: off [fixed]
> l2-fwd-offload: off [fixed]
> busy-poll: off [fixed]
> hw-tc-offload: off [fixed]

It’s tg3 indeed, super, thanks. I may have to be disabling some of those offloads, but that’s typical.

>> or similar? If so, I see BQL support in that code,
> 
> 	I believe sometime in the past I confirmed BQL working, but my memory is hazy and this might have been a different machine…

I’m sure it works, from what I see in tg3.c. Support was added in Nov. 2011 by people with experience doing this, and it uses the multi-queue interfaces also. It looks good.

https://github.com/torvalds/linux/commit/298376d3e8f00147548c426959ce79efc47b669a <https://github.com/torvalds/linux/commit/298376d3e8f00147548c426959ce79efc47b669a>

https://github.com/torvalds/linux/commit/5cb917bc4f3882ecee87064483111023086757d3 <https://github.com/torvalds/linux/commit/5cb917bc4f3882ecee87064483111023086757d3>

>> I see a Gen 10 available here for slightly more with the same BCM5720 NIC, and a AMD Opteron X3216 instead, plus 8 gigs of RAM, which I don’t need really, but I can search more for similar hardware: https://www.alza.cz/hpe-proliant-microserver-gen10-d5127723.htm?o=1 <https://www.alza.cz/hpe-proliant-microserver-gen10-d5127723.htm?o=1>
> 
> 	Then gen10 microserver while newer comes with a soldered in pre-ZEN amd cpu and seems less extensible then the gen8, but then the gen10 probly is also a decent machine.

There is a difference in TDP (35W for gen 8 and 15W for gen 10), which I might appreciate in the smaller space I have for it, otherwise the CPU benchmark differences look minimal. As is often the case, the new generation may be as much about lowering costs as anything else. The Czech retailer I use only has the gen 10 now, so I’ll probably choose it if only because of that.

I wonder if this would be enough of a machine to run both the client and server, one on each NIC, then I could use a USB NIC for management. Have you tried this with Flent? There would be no time sync issue. :) I would not be surprised if this affects the results on higher counts of streams though, or I might be missing other reasons why this is a bad idea.

I’ll go for at least one of these, but still deciding if to have an early Christmas and go for two…


[-- Attachment #2: Type: text/html, Size: 55717 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Bob McMahon]

[-- Attachment #1: Type: text/plain, Size: 4825 bytes --]

I have some brix with realtek and run ptpd installed with fedora 25.    The
corrections are in the 25 microsecond range, though there are anomalies.
These are used for wifi DUTs that go into RF enclosures.

[root@hera ~]# tail -n 1 /var/log/ptpd2.stats
2017-11-04 18:33:46.723476, slv, 0cc47afffea87386(unknown)/1,  0.000000000,
-0.000018381,  0.000000000, -0.000018463, 1528.032750001, S, 0.000000000,
0, -0.000018988, 1403, 1576, 17, -0.000018463,  0.000000000

For LAN/WAN traffic, I tend to use the intel quad server adapters in a
supermicro mb desktop with 8 or more real cores.  (I think the data center
class machines are worth it.)

Here's the brix info:

[root@hera ~]# dmidecode -t1
# dmidecode 3.1
Getting SMBIOS data from sysfs.
SMBIOS 2.7 present.

Handle 0x0001, DMI type 1, 27 bytes
System Information
Manufacturer: GIGABYTE
Product Name: MMLP3AP-00
Version: 1.x
Serial Number: To be filled by O.E.M.
UUID: 038D0240-045C-05F7-5C06-9F0700080009
Wake-up Type: Power Switch
SKU Number: To be filled by O.E.M.
Family: To be filled by O.E.M.

[root@hera ~]# dmidecode -t4
# dmidecode 3.1
Getting SMBIOS data from sysfs.
SMBIOS 2.7 present.

Handle 0x003E, DMI type 4, 42 bytes
Processor Information
Socket Designation: SOCKET 0
Type: Central Processor
Family: Core i7
Manufacturer: Intel
ID: 51 06 04 00 FF FB EB BF
Signature: Type 0, Family 6, Model 69, Stepping 1
Flags:
FPU (Floating-point unit on-chip)
VME (Virtual mode extension)
DE (Debugging extension)
PSE (Page size extension)
TSC (Time stamp counter)
MSR (Model specific registers)
PAE (Physical address extension)
MCE (Machine check exception)
CX8 (CMPXCHG8 instruction supported)
APIC (On-chip APIC hardware supported)
SEP (Fast system call)
MTRR (Memory type range registers)
PGE (Page global enable)
MCA (Machine check architecture)
CMOV (Conditional move instruction supported)
PAT (Page attribute table)
PSE-36 (36-bit page size extension)
CLFSH (CLFLUSH instruction supported)
DS (Debug store)
ACPI (ACPI supported)
MMX (MMX technology supported)
FXSR (FXSAVE and FXSTOR instructions supported)
SSE (Streaming SIMD extensions)
SSE2 (Streaming SIMD extensions 2)
SS (Self-snoop)
HTT (Multi-threading)
TM (Thermal monitor supported)
PBE (Pending break enabled)
Version: Intel(R) Core(TM) i7-4500U CPU @ 1.80GHz
Voltage: 1.2 V
External Clock: 100 MHz
Max Speed: 3800 MHz
Current Speed: 1800 MHz
Status: Populated, Enabled
Upgrade: Socket rPGA988B
L1 Cache Handle: 0x0040
L2 Cache Handle: 0x003F
L3 Cache Handle: 0x0041
Serial Number: Not Specified
Asset Tag: Fill By OEM
Part Number: Fill By OEM
Core Count: 2
Core Enabled: 2
Thread Count: 4
Characteristics:
64-bit capable


Bob

On Sat, Nov 4, 2017 at 6:33 AM, Pete Heist <peteheist@gmail.com> wrote:

> My Flent test rig needs a refresh. I currently use two Mac Minis. The
> Intel with the Core 2 Duo P7550 and forcedeth Ethernet is OK, so I could
> possibly keep it (but no BQL support which is not ideal for some tests),
> but it’s time for my G4 Mini with 100 Mbit Ethernet and stratospheric clock
> drift to go.
>
> So I’m searching for one or two low-cost Flent devices (client or server-
> which needs more CPU by the way?). Requirements:
>
> - Gigabit Ethernet (1x ok, 2x better) with a reliable Linux driver with
> BQL support
> - PTP timestamp support (http://linuxptp.sourceforge.net), hardware
> preferable
> - enough CPU to accurately do Flent’s higher flow count tests like
> rrul_torrent or rrul_be_nflows with 64 flows (sometimes I also do these
> together with one or more instances of rrul_be to test host fairness)
>
> I started a spreadsheet of what I’ve found so far here in Czech:
>
> https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_
> GVllFfiH9RU5ws5l_aY/edit?usp=sharing
>
> I’m leaning towards either one of the low-end Intel NUCs or GIGABYTE BRIX
> mini PCs, to which I’d add some memory and storage. But so many of these
> low-end devices come with Realtek Ethernet (r8169 driver?). I’m not sure
> how stable that driver is, if the BQL support is usable (
> https://www.bufferbloat.net/projects/bloat/wiki/BQL_enabled_drivers/),
> and I’d have to settle for software timestamp support for PTP. With a
> higher budget, I might go for a 1U server with well-supported Intel NICs,
> something like described in this setup (http://www.academia.edu/
> 10312557/DEVELOPING_LOW-COST_NTP_STRATUM_1_SERVERS_WITH_LINUX_PTP_AND_GPS).
> But I can’t seem to find those cheaply (even <$1000) here.
>
> Any thoughts or ideas on this?
>
>
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>

[-- Attachment #2: Type: text/html, Size: 9103 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Jonathan Morton]

[-- Attachment #1: Type: text/plain, Size: 1022 bytes --]

I'm currently using an AMD APU system on an AsRock mini-ITX board as my
primary computer.  That happened to come with Atheros network hardware,
both wired and wireless.

With a little effort, it could be squeezed into a very small case indeed,
and the major components can probably be found on clearance since the Ryzen
release.  APUs' CPU cores are hardly state of the art, but quick enough for
this sort of job and sometimes very cheap indeed.

The m/board model I have is FM2A88X-ITX+.  Note that the later "/ac" model
goes back to a Realtek wired chipset.  If you can stand a larger form
factor, the FM2A88M Extreme4+ R2.0 is another option to compare on price,
though it doesn't come with a wifi card.

In the current AsRock lineup for the AM4 socket, it appears that some
models in their Fatal1ty line use Intel NICs.  Pair one of those with a
Bristol Ridge APU for now, and you'll be able to upgrade to Zen based Raven
Ridge APUs, expected to be easily twice as fast, when they become available.

- Jonathan Morton

[-- Attachment #2: Type: text/html, Size: 1146 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 2048 bytes --]


> On Nov 5, 2017, at 1:13 PM, Jonathan Morton <chromatix99@gmail.com> wrote:
> I'm currently using an AMD APU system on an AsRock mini-ITX board as my primary computer.  That happened to come with Atheros network hardware, both wired and wireless.
> 
> With a little effort, it could be squeezed into a very small case indeed, and the major components can probably be found on clearance since the Ryzen release.  APUs' CPU cores are hardly state of the art, but quick enough for this sort of job and sometimes very cheap indeed.
> 
> The m/board model I have is FM2A88X-ITX+.  Note that the later "/ac" model goes back to a Realtek wired chipset.  If you can stand a larger form factor, the FM2A88M Extreme4+ R2.0 is another option to compare on price, though it doesn't come with a wifi card.
> 
> In the current AsRock lineup for the AM4 socket, it appears that some models in their Fatal1ty line use Intel NICs.  Pair one of those with a Bristol Ridge APU for now, and you'll be able to upgrade to Zen based Raven Ridge APUs, expected to be easily twice as fast, when they become available.
> 
Thanks for that info. As for my primary computer, I’m thinking about what I’m going to do when my MBP’s mainboard finally goes. I like the hardware and CPU, but this is its third mainboard (twice for graphics faults), they’re not even selling replacements anymore and it’s probably only a matter of time. Since right now I’m mainly only using vim and the Go compiler, I don’t even need what I have, or a new MBP with a "touch bar", so I may look for a system like this.

As far as a test box though, I think (hope) that I can get away with less hardware than this. Although, as mentioned I’m now thinking of running the Flent client and server on the same box (usually testing through point-to-point connections and router boxes), and for that, more than 2 cores would probably help. I see your point that the hardware you mentioned can be had relatively cheaply, so I’ll compare this with what I’m already looking at.

[-- Attachment #2: Type: text/html, Size: 2566 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Jonathan Morton]

[-- Attachment #1: Type: text/plain, Size: 719 bytes --]

If budget is a primary concern, there are a number of boards available with
the AM1 socket which is designed around low-power, low-cost APUs, or even
with an APU soldered to them.  AMD are phasing that platform out right now,
but it's still available since there isn't a direct replacement at the same
price yet.

What you'll get here are the small "cat series" cores designed for netbooks
and game consoles, but a quad-core version remains pretty cheap.  My E-450
is an older, dual-core version of this and is capable enough, like a
slightly gimped C2D.

However, I haven't yet found one with anything other than a bog standard
Realtek NIC.  They do have PCIe slots, so you can add a discrete card.

- Jonathan Morton

[-- Attachment #2: Type: text/html, Size: 824 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 2126 bytes --]


> On Nov 5, 2017, at 5:24 PM, Jonathan Morton <chromatix99@gmail.com> wrote:
> If budget is a primary concern, there are a number of boards available with the AM1 socket which is designed around low-power, low-cost APUs, or even with an APU soldered to them.  AMD are phasing that platform out right now, but it's still available since there isn't a direct replacement at the same price yet.
> 
> What you'll get here are the small "cat series" cores designed for netbooks and game consoles, but a quad-core version remains pretty cheap.  My E-450 is an older, dual-core version of this and is capable enough, like a slightly gimped C2D.
> 
> However, I haven't yet found one with anything other than a bog standard Realtek NIC.  They do have PCIe slots, so you can add a discrete card.
> 
From this info I spec’d out two builds and compared it with the Proliant Microserver: https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=199471438 <https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=199471438>

You’re right, the toughest part with low-end hardware is finding high quality and/or dual-port NICs, which is what seems to make the Proliant a pretty good deal. Most hardware in this range has Realteks, which might be ok, but I’m not convinced, so I added a dual port Intel PRO/1000 to each of my two builds, which then makes their price a bit higher than the Proliant.

One advantage to the external PRO/1000 is that it supports PCIe x4 whereas the BCM5720 that comes with the Proliant only supports PCIe x2, so there’s a bit of a difference here, which I may or may not notice. On the other hand, the BCM5720 supports hardware timestamping whereas the Intel does not.

I guess I should focus on the big picture here, either quad-core is important vs dual-core or it isn’t, and either I can do my testing with one box for both client and server or I need two. Neither I’m entirely sure about, so I might end up ordering something with the option to return so I can run some tests.

Thanks again for the tips...

[-- Attachment #2: Type: text/html, Size: 2791 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Jonathan Morton]

[-- Attachment #1: Type: text/plain, Size: 269 bytes --]

I see you've specced them to have PicoPSUs and so on.  That isn't exactly
the cheapest option, though it is very compact - which I know from
experience since my E-450 uses the same.  Considering less compact options
may improve the price comparison.

- Jonathan Morton

[-- Attachment #2: Type: text/html, Size: 321 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 1105 bytes --]


> On Nov 5, 2017, at 7:08 PM, Jonathan Morton <chromatix99@gmail.com> wrote:
> I see you've specced them to have PicoPSUs and so on.  That isn't exactly the cheapest option, though it is very compact - which I know from experience since my E-450 uses the same.  Considering less compact options may improve the price comparison.
> 
It's true, but also if I wanted to compare my builds to the ProLiant fairly, they’d have to include 8GB of RAM. So I put up four more builds, two with the cheapest case and PSU I could find on the same site, and two with the same cheap case/PSU but also 8GB of RAM.

The prices are all close enough for me to where I’d just want to choose what’s best. The single-threaded performance of the quad-core CPUs are naturally lower, which may affect something, somehow. It would be nice to be able to try Flent tests on both without buying both, but that’s how it is.

Also, with any build it takes a little more time and there’s the possibility of a compatibility problem vs buying something pre-tested (lessons learned), but these builds do look pretty safe.

[-- Attachment #2: Type: text/html, Size: 1603 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Jonathan Morton]

[-- Attachment #1: Type: text/plain, Size: 1007 bytes --]

Regarding single-core performance, the Sempron 3850 may be better in that
respect than my E-450, which itself is perfectly capable of handling GigE
at line rate (unlike an Atom).  It has an improved core design which may
compensate for the lower clock speed.

There are three members of AMD's "cat" family: Bobcat, Jaguar, Puma.  My
E-450 has Bobcat, which is the oldest and maxes out at 1.65 GHz (there is
no turbo clock).  Various consoles got Jaguar cores, as did the Kabini APU
which forms your Sempron.  The Puma core is the newest, and appears in
"Carrizo-L" APUs in last year's netbooks.  The design improvements between
Bobcat and Jaguar are much larger than between Jaguar and Puma; consult
Agner Fog's optimisation manuals if you want to know the gory details.

The AM1 Kabini family also includes the Athlon 5150 and 5350, which are
exactly the same as the 3850 except for clock speeds.  If the 3850 isn't
fast enough, you can get an easy 50% boost from upgrading to the 5350.

- Jonathan Morton

[-- Attachment #2: Type: text/html, Size: 1125 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 3708 bytes --]


> On Nov 5, 2017, at 9:15 PM, Jonathan Morton <chromatix99@gmail.com> wrote:
> Regarding single-core performance, the Sempron 3850 may be better in that respect than my E-450, which itself is perfectly capable of handling GigE at line rate (unlike an Atom).  It has an improved core design which may compensate for the lower clock speed.
> 
> There are three members of AMD's "cat" family: Bobcat, Jaguar, Puma.  My E-450 has Bobcat, which is the oldest and maxes out at 1.65 GHz (there is no turbo clock).  Various consoles got Jaguar cores, as did the Kabini APU which forms your Sempron.  The Puma core is the newest, and appears in "Carrizo-L" APUs in last year's netbooks.  The design improvements between Bobcat and Jaguar are much larger than between Jaguar and Puma; consult Agner Fog's optimisation manuals if you want to know the gory details.
> 
> The AM1 Kabini family also includes the Athlon 5150 and 5350, which are exactly the same as the 3850 except for clock speeds.  If the 3850 isn't fast enough, you can get an easy 50% boost from upgrading to the 5350.
> 

These days it seems like most hardware is capable of GigE at line rate, but the smoothness of individual flows or other factors may be affected by CPU, either cores or speed. I wrote a script to run rrul_be_nflows over the loopback adapter and vary a few things:

- number of flows
- number of CPU cores enabled
- CPU frequency
- whether or not to limit loopback to 2gbit (simulating GigE)

This was an attempt to learn more about how the number of cores and CPU frequency can affect the results, but because I don’t actually have a box with two NICs and this uses loopback, it's just for learning, and should be taken with a grain of salt. Here are a few of the results from my MBP and Mini (six tabs of results):

https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=739582898 <https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=739582898>

Not too many surprises:
- I saw more dramatic differences overall when going from 1-2 cores (not good to test with only a single core)
- other things being equal, more cores are capable of higher throughput, but > 4 cores didn’t change much on my MBP
- with higher flow counts, more cores can make individual flows a bit smoother, and ping times sometimes lower and smoother
- I’d rather have a dual-core 1.6 GHz Core 2 Duo than a single-core 2.26 GHz Core 2 Duo, but because my hardware doesn’t let me lower CPU frequency to half of maximum, I couldn’t test a doubling of frequency vs going from 1-2 cores, which would be interesting to me (anyone else with hardware that can do that care to try?)

If anyone wants to run the script (takes a little modification), it’s here:

https://drive.google.com/file/d/1oasB3ePDMMr1WdRi_8koQmRO_qUjVB52/view?usp=sharing <https://drive.google.com/file/d/1oasB3ePDMMr1WdRi_8koQmRO_qUjVB52/view?usp=sharing>

As for the hardware, running everything on one box looks more complicated than I thought, not just because of Flent, but because of the scripts I wrote for setup and teardown and what they do. So now I’m thinking of taking one for the team and getting two boxes to avoid the hassle. I’ll still see.

Anyway, I added some ASUS / Intel builds for more on the high end:

https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=199471438 <https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=199471438>

If I ever see a comparison between the effects of doubling clock speed or going from 2-4 cores, it might help decide on the CPU.

[-- Attachment #2: Type: text/html, Size: 4856 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 2256 bytes --]


> On Nov 5, 2017, at 2:42 AM, Bob McMahon <bob.mcmahon@broadcom.com> wrote:
> 
> I have some brix with realtek and run ptpd installed with fedora 25.    The corrections are in the 25 microsecond range, though there are anomalies.  These are used for wifi DUTs that go into RF enclosures.  
> 
> [root@hera ~]# tail -n 1 /var/log/ptpd2.stats
> 2017-11-04 18:33:46.723476, slv, 0cc47afffea87386(unknown)/1,  0.000000000, -0.000018381,  0.000000000, -0.000018463, 1528.032750001, S, 0.000000000, 0, -0.000018988, 1403, 1576, 17, -0.000018463,  0.000000000
> 
> For LAN/WAN traffic, I tend to use the intel quad server adapters in a supermicro mb desktop with 8 or more real cores.  (I think the data center class machines are worth it.)

Thanks for the info. I was wondering how large the PTP error would be with software timestamps, and I see it’s not bad for most purposes.

Which Realtek Linux driver does your brix use, and is it stable? The r8169 driver’s BQL support was reverted at some point and it doesn’t look like that has changed.

I trust that the extra cores can help, particularly for tests with high flow counts, but my project budget won’t allow it, and used hardware is too much to think about at the moment.

Do you (or anyone) know of any problems with running the Flent client and server on the same box? In the case of the Proliant Microserver, the Broadcom 5720 adapter should have separate PCI data paths for each NIC. I guess the bottleneck will still mainly be the CPU. To get some idea of what's possible on my current hardware, I tried running rrul_be_nflows tests with the Flent client and server on the same box, through its local adapter (with MTU set to 1500) with my current Mac Mini (2.26 GHz Core2 Duo P7550). I know that doesn’t predict how it will work over Ethernet, but it’s a start.

https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=1583696271 <https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=1583696271>

Although total throughput is pretty good for a low-end CPU, I’m not sure I’d trust the results above 64/64 flows. 256/256 flows was an epic fail, but I won’t be doing that kind of test.

[-- Attachment #2: Type: text/html, Size: 3113 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Bob McMahon]

[-- Attachment #1: Type: text/plain, Size: 2857 bytes --]

The realtek is definitely not ideal.   The test needs to watch the ptp
stats to make sure the corrections are stable over the life of the test and
throw out bad results per test equipment introducing too much error.

Qualifying a NIC for use in test equipment is a bit of a pain.  My
rationale is to avoid consumer grade products, rather leverage the work of
engineers that qualify equipment for data centers, i.e. the data center
market is driving the vendor.   I find the INTC server class NICs to be the
best for this so far.

Bob

On Sun, Nov 5, 2017 at 5:57 AM, Pete Heist <peteheist@gmail.com> wrote:

>
> On Nov 5, 2017, at 2:42 AM, Bob McMahon <bob.mcmahon@broadcom.com> wrote:
>
> I have some brix with realtek and run ptpd installed with fedora 25.
> The corrections are in the 25 microsecond range, though there are
> anomalies.  These are used for wifi DUTs that go into RF enclosures.
>
> [root@hera ~]# tail -n 1 /var/log/ptpd2.stats
> 2017-11-04 18:33:46.723476, slv, 0cc47afffea87386(unknown)/1,
> 0.000000000, -0.000018381,  0.000000000, -0.000018463, 1528.032750001, S,
> 0.000000000, 0, -0.000018988, 1403, 1576, 17, -0.000018463,  0.000000000
>
> For LAN/WAN traffic, I tend to use the intel quad server adapters in a
> supermicro mb desktop with 8 or more real cores.  (I think the data center
> class machines are worth it.)
>
>
> Thanks for the info. I was wondering how large the PTP error would be with
> software timestamps, and I see it’s not bad for most purposes.
>
> Which Realtek Linux driver does your brix use, and is it stable? The r8169
> driver’s BQL support was reverted at some point and it doesn’t look like
> that has changed.
>
> I trust that the extra cores can help, particularly for tests with high
> flow counts, but my project budget won’t allow it, and used hardware is too
> much to think about at the moment.
>
> Do you (or anyone) know of any problems with running the Flent client and
> server on the same box? In the case of the Proliant Microserver, the
> Broadcom 5720 adapter should have separate PCI data paths for each NIC. I
> guess the bottleneck will still mainly be the CPU. To get some idea of
> what's possible on my current hardware, I tried running rrul_be_nflows
> tests with the Flent client and server on the same box, through its local
> adapter (with MTU set to 1500) with my current Mac Mini (2.26 GHz Core2 Duo
> P7550). I know that doesn’t predict how it will work over Ethernet, but
> it’s a start.
>
> https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_
> GVllFfiH9RU5ws5l_aY/edit#gid=1583696271
>
> Although total throughput is pretty good for a low-end CPU, I’m not sure
> I’d trust the results above 64/64 flows. 256/256 flows was an epic fail,
> but I won’t be doing that kind of test.
>

[-- Attachment #2: Type: text/html, Size: 3734 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 1319 bytes --]


> On Nov 5, 2017, at 8:42 PM, Bob McMahon <bob.mcmahon@broadcom.com> wrote:
> 
> The realtek is definitely not ideal.   The test needs to watch the ptp stats to make sure the corrections are stable over the life of the test and throw out bad results per test equipment introducing too much error.
> 
> Qualifying a NIC for use in test equipment is a bit of a pain.  My rationale is to avoid consumer grade products, rather leverage the work of engineers that qualify equipment for data centers, i.e. the data center market is driving the vendor.   I find the INTC server class NICs to be the best for this so far.

I figured as much, and might take it into account that you’re steering me towards those NICs. I’m pretty sure the I350T2V2 would work well, so I started another build with it, despite the price:

https://www.senetic.cz/product/I350T2V2?gclid=EAIaIQobChMI6u7YkKKo1wIVqwrTCh0anguMEAAYASAAEgK3O_D_BwE <https://www.senetic.cz/product/I350T2V2?gclid=EAIaIQobChMI6u7YkKKo1wIVqwrTCh0anguMEAAYASAAEgK3O_D_BwE>

My latest thought is, if I get a good dual port NIC and go at least a bit higher-end with the CPU, I shouldn't need a second device, and time synchronization won’t be an issue, so it could save some effort. I’ll try a few more options, sleep on it and thanks everyone for the help!

[-- Attachment #2: Type: text/html, Size: 1980 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Isaac Konikoff]

[-- Attachment #1: Type: text/plain, Size: 3189 bytes --]

You can run flent/iperf/netperf client and server on the same box using a
candelatech kernel and then bind to specific interfaces.

http://candelatech.com/private/downloads/r5.3.6/ct4.9.29+.x64.tar.gz
guest/guest

flent example:
eth1 192.168.1.2 to DUT(AP LAN side)
wlan0 192.168.1.3 to DUT(AP wireless)

netserver
flent -H 192.168.1.3 --local-bind 192.168.1.2 --swap-up-down -x
tcp_download -l 120


iperf example:
eth1 192.168.86.103
wlan0 192.168.86.101

iperf upload test
iperf -s -B 192.168.86.103 -i10
iperf -c 192.168.86.103 -B 192.168.86.101 -i10 -t120

iperf download test
iperf -s -B 192.168.86.101 -i10
iperf -c 192.168.86.101 -B 192.168.86.103 -i10 -t120


On Sun, Nov 5, 2017 at 5:57 AM, Pete Heist <peteheist@gmail.com> wrote:

>
> On Nov 5, 2017, at 2:42 AM, Bob McMahon <bob.mcmahon@broadcom.com> wrote:
>
> I have some brix with realtek and run ptpd installed with fedora 25.
> The corrections are in the 25 microsecond range, though there are
> anomalies.  These are used for wifi DUTs that go into RF enclosures.
>
> [root@hera ~]# tail -n 1 /var/log/ptpd2.stats
> 2017-11-04 18:33:46.723476, slv, 0cc47afffea87386(unknown)/1,
> 0.000000000, -0.000018381,  0.000000000, -0.000018463, 1528.032750001, S,
> 0.000000000, 0, -0.000018988, 1403, 1576, 17, -0.000018463,  0.000000000
>
> For LAN/WAN traffic, I tend to use the intel quad server adapters in a
> supermicro mb desktop with 8 or more real cores.  (I think the data center
> class machines are worth it.)
>
>
> Thanks for the info. I was wondering how large the PTP error would be with
> software timestamps, and I see it’s not bad for most purposes.
>
> Which Realtek Linux driver does your brix use, and is it stable? The r8169
> driver’s BQL support was reverted at some point and it doesn’t look like
> that has changed.
>
> I trust that the extra cores can help, particularly for tests with high
> flow counts, but my project budget won’t allow it, and used hardware is too
> much to think about at the moment.
>
> Do you (or anyone) know of any problems with running the Flent client and
> server on the same box? In the case of the Proliant Microserver, the
> Broadcom 5720 adapter should have separate PCI data paths for each NIC. I
> guess the bottleneck will still mainly be the CPU. To get some idea of
> what's possible on my current hardware, I tried running rrul_be_nflows
> tests with the Flent client and server on the same box, through its local
> adapter (with MTU set to 1500) with my current Mac Mini (2.26 GHz Core2 Duo
> P7550). I know that doesn’t predict how it will work over Ethernet, but
> it’s a start.
>
> https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_
> GVllFfiH9RU5ws5l_aY/edit#gid=1583696271
>
> Although total throughput is pretty good for a low-end CPU, I’m not sure
> I’d trust the results above 64/64 flows. 256/256 flows was an epic fail,
> but I won’t be doing that kind of test.
>
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>

[-- Attachment #2: Type: text/html, Size: 4579 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Bob McMahon]

[-- Attachment #1: Type: text/plain, Size: 3708 bytes --]

Just to be clear, speaking for iperf 2, the binding isn't to an interface
but to an IP address.   See this for a description
<https://sourceforge.net/p/iperf2/discussion/general/thread/f856ae2c/>.

Linux supports SO_BINDTODEVICE but it's not straightforward per things like
ARP so I didn't add support for this.

Bob

On Mon, Nov 6, 2017 at 12:15 PM, Isaac Konikoff <isaac.konikoff@gmail.com>
wrote:

> You can run flent/iperf/netperf client and server on the same box using a
> candelatech kernel and then bind to specific interfaces.
>
> http://candelatech.com/private/downloads/r5.3.6/ct4.9.29+.x64.tar.gz
> guest/guest
>
> flent example:
> eth1 192.168.1.2 to DUT(AP LAN side)
> wlan0 192.168.1.3 to DUT(AP wireless)
>
> netserver
> flent -H 192.168.1.3 --local-bind 192.168.1.2 --swap-up-down -x
> tcp_download -l 120
>
>
> iperf example:
> eth1 192.168.86.103
> wlan0 192.168.86.101
>
> iperf upload test
> iperf -s -B 192.168.86.103 -i10
> iperf -c 192.168.86.103 -B 192.168.86.101 -i10 -t120
>
> iperf download test
> iperf -s -B 192.168.86.101 -i10
> iperf -c 192.168.86.101 -B 192.168.86.103 -i10 -t120
>
>
> On Sun, Nov 5, 2017 at 5:57 AM, Pete Heist <peteheist@gmail.com> wrote:
>
>>
>> On Nov 5, 2017, at 2:42 AM, Bob McMahon <bob.mcmahon@broadcom.com> wrote:
>>
>> I have some brix with realtek and run ptpd installed with fedora 25.
>> The corrections are in the 25 microsecond range, though there are
>> anomalies.  These are used for wifi DUTs that go into RF enclosures.
>>
>> [root@hera ~]# tail -n 1 /var/log/ptpd2.stats
>> 2017-11-04 18:33:46.723476, slv, 0cc47afffea87386(unknown)/1,
>> 0.000000000, -0.000018381,  0.000000000, -0.000018463, 1528.032750001,
>> S, 0.000000000, 0, -0.000018988, 1403, 1576, 17, -0.000018463,  0.000000000
>>
>> For LAN/WAN traffic, I tend to use the intel quad server adapters in a
>> supermicro mb desktop with 8 or more real cores.  (I think the data center
>> class machines are worth it.)
>>
>>
>> Thanks for the info. I was wondering how large the PTP error would be
>> with software timestamps, and I see it’s not bad for most purposes.
>>
>> Which Realtek Linux driver does your brix use, and is it stable? The
>> r8169 driver’s BQL support was reverted at some point and it doesn’t look
>> like that has changed.
>>
>> I trust that the extra cores can help, particularly for tests with high
>> flow counts, but my project budget won’t allow it, and used hardware is too
>> much to think about at the moment.
>>
>> Do you (or anyone) know of any problems with running the Flent client and
>> server on the same box? In the case of the Proliant Microserver, the
>> Broadcom 5720 adapter should have separate PCI data paths for each NIC. I
>> guess the bottleneck will still mainly be the CPU. To get some idea of
>> what's possible on my current hardware, I tried running rrul_be_nflows
>> tests with the Flent client and server on the same box, through its local
>> adapter (with MTU set to 1500) with my current Mac Mini (2.26 GHz Core2 Duo
>> P7550). I know that doesn’t predict how it will work over Ethernet, but
>> it’s a start.
>>
>> https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheN
>> FrH_GVllFfiH9RU5ws5l_aY/edit#gid=1583696271
>>
>> Although total throughput is pretty good for a low-end CPU, I’m not sure
>> I’d trust the results above 64/64 flows. 256/256 flows was an epic fail,
>> but I won’t be doing that kind of test.
>>
>> _______________________________________________
>> Make-wifi-fast mailing list
>> Make-wifi-fast@lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>>
>
>

[-- Attachment #2: Type: text/html, Size: 5456 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Dave Taht]

I was away from keyboard for much of this thread, I wanted to point out
that I find the pcengines apu2 series very useful as routers and
testbeds. They are way easier to open up than the nucs, have 3 intel
ethernet ports, AND 2-3 mini-pcie ports, and can be had, with memory for
under 200 bucks.

That said, the nucs are often convienent because they have video chips.

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 1337 bytes --]


> On Nov 7, 2017, at 1:39 AM, Dave Taht <dave@taht.net> wrote:
> 
> I was away from keyboard for much of this thread, I wanted to point out
> that I find the pcengines apu2 series very useful as routers and
> testbeds. They are way easier to open up than the nucs, have 3 intel
> ethernet ports, AND 2-3 mini-pcie ports, and can be had, with memory for
> under 200 bucks.
> 
> That said, the nucs are often convienent because they have video chips.

Well _that’s_ interesting, I actually already have 2x APU (not APU2) from FreeNet (my ISP) for testing. FreeNet uses these as routers.

But, the APUs I have are V1 and have a dual core with Realtek 8111E, where the APU2 has a quad core with Intel i210/211AT. The APU2 looks far preferable for testing, and I hadn’t noticed its existence. It’s dramatically cheaper than anything else I was looking at, and since I don’t need video, if the CPUs are fast enough, they look very interesting as a way to save a lot of money.

So I added those as a build:

https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=199471438 <https://docs.google.com/spreadsheets/d/1MVxGsreiGKNXhfkMIheNFrH_GVllFfiH9RU5ws5l_aY/edit#gid=199471438>

Basically I could get two of those for one of most anything I was looking at. Thanks for that idea!

[-- Attachment #2: Type: text/html, Size: 2041 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Pete Heist]

[-- Attachment #1: Type: text/plain, Size: 885 bytes --]

I ordered 2x APU2: http://pcengines.ch/apu2c4.htm <http://pcengines.ch/apu2c4.htm>

With 16GB mSATA SSDs, PSUs, cases and the serial cable I needed, the total came to 9097 CZK / 356 EUR / 412 USD.

Happily, the igb Ethernet driver for its i210 NICs does support BQL and appears to have hardware timestamping to boot. With three Ethernet ports, I’ll get one for management, one for testing and one for PTP time sync straight to the other APU.

I’m expecting the CPU PassMark (for what that’s worth) to be around 1100, 353 single core. That’s significantly less than other devices I was looking at, but at least it’s a quad core, and at this price it’s worth a try. Since Jon reported his E-450 (PassMark 769, 431 single core) was capable of line-rate GigE, hopefully this will be as well.

When they come I’ll run some benchmarks and report back how it goes…


[-- Attachment #2: Type: text/html, Size: 1366 bytes --]

Re: [Make-wifi-fast] Flent test hardware

[Bob McMahon]

[-- Attachment #1: Type: text/plain, Size: 1830 bytes --]

On conducted topologies, here are two common ones where ln are variable
attenuator legs.  The asterisk is a butler matrix or splitter/combiner in
the relative center.

A       B
  \     /
l1\   / l2
    \ /
     *
    / \
l3/   \l4
  /     \
C      D

Then the power levels are controlled by variable attenuator legs, giving 4
degrees of freedom.

A       B
  \     /
l1\   / l2
    \ /
     *
     | l5
     *
    / \
l3/   \l4
  /     \
C      D

This immediate above collapses to the star when l5 = 0.  Otherwise l5
separates two BSSIDs as an example.

The matrices are simple, eg. for the star:

1 1 0 0
1 0 1 0
0 0 1 1
0 1 0 1

Bob

On Tue, Nov 7, 2017 at 2:21 AM, Pete Heist <peteheist@gmail.com> wrote:

> I ordered 2x APU2: http://pcengines.ch/apu2c4.htm
>
> With 16GB mSATA SSDs, PSUs, cases and the serial cable I needed, the total
> came to 9097 CZK / 356 EUR / 412 USD.
>
> Happily, the igb Ethernet driver for its i210 NICs does support BQL and
> appears to have hardware timestamping to boot. With three Ethernet ports,
> I’ll get one for management, one for testing and one for PTP time sync
> straight to the other APU.
>
> I’m expecting the CPU PassMark (for what that’s worth) to be around 1100,
> 353 single core. That’s significantly less than other devices I was looking
> at, but at least it’s a quad core, and at this price it’s worth a try.
> Since Jon reported his E-450 (PassMark 769, 431 single core) was capable of
> line-rate GigE, hopefully this will be as well.
>
> When they come I’ll run some benchmarks and report back how it goes…
>
>
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>

[-- Attachment #2: Type: text/html, Size: 2840 bytes --]