V3 for x86PC nearly ready

[lonnie]

Just a heads up that I upgraded several PC Routers today to V3 and they both came up and work great. Expect a release for the PC Architecture in several days. The Server is not happening with radius and squid, but Router with Atheros support looks to be fine. We have tried Intel chipset for P4 and VIA chipset for dual P3. Dual cpu might be supported. We'll see what happens tomorrow.

As a note, the FREE version will be useful as a Router and has full quagga Ethernet support.

The paid version will have the same features as the WAR and WRAP versions, except some of the systems will not have a hardware watchdog. The Intel mobo has hardware watchdog, so we are pleased. It seems that 16 MB is full ram use compared to 45 MB in v2. This new stuff is really a step forward in performance and a step backward in resource usage.

I still prefer the quad WAR for simplicity and low power draw, but the P4 sure rocks.

[mp3turbo]

this is the most amazing product news this year [although you don't see too much market for x86]. p3/p4 platform has superior performance to all those 0.2W WARs :) and there's much use of it where you have 110V available.

Lonnie, can you elaborate a bit on x86 wireless performance compared to v2 ?

[ninedd]

This is GREAT news for me. This is the big key to go V3 everywhere for me. :) I have quite a few building top locations, with PC's in utility / maintenance rooms / etc. I wasn't looking forward to spending the loot to replace them with WAR's. Even though it would improve our infrastucture, it would have been replacing already working hardware and not actually increased our infrastructure. This is REALLY good news - now I can do both, go V3 everywhere, keep our working hardware, and put WAR's in NEW locations as well. Did I mention this is good news? :) Bang on guys!

[mp3turbo]

exactly the same applies to my situation, inet2000 ! Lonnie, another couple of questions came into my mind:


a) does l7 filtering catch encrypted bittorent? Is is possible to shape encrypted traffic or drop only? Anyone with experience... ?

b) I hope to see "clients are now much more resistant to interference" and "atheros transmit retry is now more aggressive" implemented in v3/x86 also :)

c) generally, I'd like to see all atheros improvements implemented there

d) I don't know how connection tracking table is handled in V2/V3 - please do not set fixed size, but calculate appropriate number of entries acc. to available RAM on each reboot. We have run into a brick wall with 32768 entries. Is size of conntrack configurable? Is there some automatic calculation which sets number of entries? Is it fixed - how much?



thnx, mp3turbo.

[tony]

a.) The latest layer-7 patterns are used, so it's just a matter of small test to see if it does what you need.

b.) All v3 features are available in x86 releases

c.) Same as b.

d.) This is tunable in the same manner as v2.

--------

Other things to note:

The v3 and v2 offerings are vastly different in terms of hardware support, and ethernet detection ordering. We cannot guerentee every device in v2 will be supported in v3, at least not right away.

Likewise, there will be ethernet and hardware platforms in v3 that are not supported in v2.

Upgrading from v2 to v3 will be avilable in the same manner as the WRAP release once ready.

[Stratolinks]

Sorry, maybe I misunderstood, but I thought that you were not going to do a generic X86 release with wireless capabilities. And here I just finished buying MT licenses for 8 of our backbone links that are on x86 industrial computer boards. And we are already seeing problems with the MT corrupting packets despite the fact they insisted this problem was fixed. You have to enter firewall mangle rules to change the tcp mss to 1452 just to make them work close to properly. The MT winbox interface looks pretty and is easy to use, but I would much rather have software that works, namely Star-OS.

Yes I'm the one who said that I wouldn't use MT for an AP even if it was free. The star-OS is way better for an AP. Actually it is way better for the backbone routers as well.

Lonnie you mentioned that the Intel boards you are testing have a watchdog. Which one, is it a variant of the port 443 or something different? If it is the port 443 type that will weigh very heavily on my switch back to Star-OS V3 on x86.

We are just getting ready to do another Bell tower install. I will be getting a quad WAR board from you for this one for the AP, which will then connect via Ethernet to our x86 router for the 5Gig backhaul. Since this is new coverage area, we don't have to worry about compatibility with existing customers, so this is going to be WLM54 cards to start with, I am going to put up one of the new Pacific Wireless sectorised omni antennas as well (I'll let everyone know in a month or 2 what my opinion of it is).

[mp3turbo]

Good morning, Tony,

a) fantastic
b) marvellous
c) wonderful
d) gorgeous


as usually, guys, you simply rock. I'm speechless.





several wishes:

- good hardware support for wireless cards (compex 54g, cm9, cm10, ar5006 chipsets); this is expected :)

- excellent support for ethernet cards (intel mainly - they have UNIFIED drivers, please use latest of them and instantly you have support for all 1Gbit metallic/fiber adapters):

http://support.intel.com/support/network/sb/cs-016238.htm

gigabit adapters drivers:
http://downloadfinder.intel.com/scripts-df-external/Detail_Desc.aspx?strState=LIVE&ProductID=983&DwnldID=9180&lang=eng
100mbit adapters: http://downloadfinder.intel.com/scripts-df-external/Detail_Desc.aspx?agr=Y&Inst=Yes&ProductID=581&DwnldID=2896&strOSs=39&OSFullName=Linux*&lang=eng

after compilation, e1000.o has around 110kB I think... small enough, universal enough. Perfect. Please use latest driver, Intel know what they are doing.

the last things I'm eagerly waiting for are:
- saturating atheros, uncompressable 11000kB/s thru wireless is my dream so far.
- ahh yes, frequency freedom. Support for ALL frequencies in ## setting, including turbo. Fully open, "all is all" including 2312-2372, 4900-6100.
- bonding support. Leave this as experimental without any support [manually editable configuration script, parameters just passed to bonding.o driver]. I hope that your atheros driver is ethernet like. You know where I'm heading to, yes I want to bond two wireless adapters.


I have to see this beast. Don't you want beta-tester with sustained 40Mbit traffic spiking into high 70ties (simplex)?

[lonnie]

Intel gigabit are built in for ethernet drivers. Intel does know what they are doing which is why the WAR boards are so good for such a small package.

We would prefer to NOT have support for 3Com.

Bonding will work on any device, even Atheros. It'll happen soon. The planned features will be multi device bonding half duplex, dual device full duplex and of course failover.

[mp3turbo]

I'm eagerly waiting for THE bonding. Huh, I really want to see it in action!

[greg]

Policy Routing included?

[tony]

All v3 features including policy routing will be available.

[greg]

Goodbye Cisco!! I wish I'd known about this sooner too. I thought the x86 platform was dead as far as v3 went.

If you need a tester - lemme know!

[Stratolinks]

I thought the x86 platform was dead as far as v3 went.
That is what I had understood too. I only wish the announcement had been made about 2 weeks sooner, it would have saved me nearly $400.00 in upgraded MT licenses on July 25th as well as several trips up silos to swap out flash cards. That's $400 that I would have gladly waited and paid to VNC (and then some) for V3 on X86 had it been made clear that this was still coming down the pipe. I didn't need an exact timeline, just confirmation that this was happening. This will add a further delay in putting V3 back in the system once it is released since the OS upgrade budget just had a good bite taken out of it and given to MT.

[gregclemmons]

Glad to hear this !

[lonnie]

Since the x86 WRAP was done we were pretty much focused on other things, but we have delayed the CPE coding a few more days and produced an image for the more standard and solid motherboard chipsets.

We did not see the need for x86PC since it had been such an unreliable platform, but we were reminded that our ATX-733 systems had been stellar performers and very reliable. The case was made that there are good motherboards and we should not give up in disgust because of some poor motherboards. So we have chosen the chipsets of some motherboards we know to be solid and we will see where it goes.

Our dual 600 MHz P3 has 1 day and 2 hours of uptime and is using the VIA chipset. Our 2.4 GHz P4 has about the same time, and uses Intel motherboard chipsets. Neither of these systems have radios and are used strictly for control and routing. The P4 was tested with a CM9 and you don't want to know what it had for throughput. It was insane.

[mp3turbo]

> The P4 was tested with a CM9 and you don't want to know what it had for throughput. It was insane.


my days are dark. Don't make me wail and whimper.

Please.



You have demonstrated something like 9000kB/s with WARs if I remember correctly, and P4 platform should be higher because of better pci and bus(es) implementation, lower wait-states (higher saturation) due to insane clocks compared to 500MHz of WARs and generally higher raw power of everything (cpu included). We are eagerly waiting to squeeze out every possible bit of performance from wireless due to our internal network DC++ systems. Even 10Mbit/s makes huge difference for us : that's 108GB more per day, 3.35TB monthly. That's why we want bonding and gigabit ethernets - our traffic is far from equally balanced in time, so half-duplex bonding will perform miracles for us.


Back to track : what was THE number, please?

[skyclimber]

I'm happy to see good news for X86 platform.

As i needed a mail and dns server on my network to avoid unecessary traffic for local mail i waited for the X86 release. Is there future plans on integrating such services (DNS with master zone records, Mail server with Webmail? Or someone have a DISTRO with WEB GUI who integrate these features.

Thanks,

Louis

[lonnie]

V3 for x86PC will be delayed until the middle of next week. We want to make sure we get it right to avoid site visits and flash replacements. As can be expected this platform has more things that can be different and those subtle changes can throw your system out of whack and require a site visit upon upgrade.

I can't guarantee it will be perfect but we will try and catch most of the things that can go wrong with the update.

[Stratolinks]

This is turning out to be the best bit of news I have heard for a while. The V3 on WRAP upgrades are still pending with more testing with all the different client hardware we have in some areas, but the backbone is where we don't have to wait, all the hardware is ready to fly with V3.

I will have to make site visits to most of the sites on our backbone now anyway to pull the MT flash card and install the V3 flash card. I still have the V2 flash card from each of them here so I can hopefully plug it in to a machine on the bench and do the upgrades here, then take them out and plug them in pre-configured. I will start the process with the sites that are easiest to access, if everything goes as smoothly as your stuff has in the past we will go ahead and do the more difficult sites pretty quickly too.

All our X86 PC routers are Intel chipsets. The newer ones are 815e with 1.1GHz 256k cache Celerons, the older ones are 440BX with 1.1GHz 128k cache Celerons.

With V2 on a 1km 5GigTurbo link we were playing with for a while, (it's not running turbo anymore) the built in TCP throughput test running from the router that has the CM9 in it to the next one with the other CM9 in it we had speeds of 1100KB/sec reported. No this is not real world data, but it does give an idea of the raw throughput that can be achieved with a bit of horsepower behind it.

So it looks like I can say goodbye to MT once and for all on the routers. Although I still have one just sitting at the head end counting all the traffic for each and every IP.

[nickwhite]

With V2 on a 1km 5GigTurbo link we were playing with for a while, (it's not running turbo anymore) the built in TCP throughput test running from the router that has the CM9 in it to the next one with the other CM9 in it we had speeds of 1100KB/sec reported. Did I miss something? Did you forget a zero(0) on the end? This is a PC-link and 1 km? I would expect more than 1100 KB/s. We're seeing 5000+ KB/s on our WAR4s and 2000 KB/s on our V2 WRAPs.

:)

Nick

[Stratolinks]

Did I miss something? Did you forget a zero(0) on the end? This is a PC-link and 1 km? I would expect more than 1100 KB/s. We're seeing 5000+ KB/s on our WAR4s and 2000 KB/s on our V2 WRAPs.

:)

Nick

OOPS! yes, I missed a zero. The results actually varied a bit betwen 10400 and 11850 I just rounded it to 11000.

The key thing was it was actually faster than running the same test between two routers connected to an old 100baseT hub. Not as fast however as plugging in to a switch that can provide full duplex on the 100Mbit ports.

[tony]

Thank you. The e100 and e1000 drivers as provided from Intel will be used.

- excellent support for ethernet cards (intel mainly - they have UNIFIED drivers, please use latest of them and instantly you have support for all 1Gbit metallic/fiber adapters):

http://support.intel.com/support/network/sb/cs-016238.htm

gigabit adapters drivers:
http://downloadfinder.intel.com/scripts-df-external/Detail_Desc.aspx?strState=LIVE&ProductID=983&DwnldID=9180&lang=eng
100mbit adapters: http://downloadfinder.intel.com/scripts-df-external/Detail_Desc.aspx?agr=Y&Inst=Yes&ProductID=581&DwnldID=2896&strOSs=39&OSFullName=Linux*&lang=eng

after compilation, e1000.o has around 110kB I think... small enough, universal enough. Perfect. Please use latest driver, Intel know what they are doing.

the last things I'm eagerly waiting for are:
- saturating atheros, uncompressable 11000kB/s thru wireless is my dream so far.

[bphantom]

Thank you. The e100 and e1000 drivers as provided from Intel will be used.

Tony/Lonnie, this is great news! Can you please test and make sure VLAN interfaces are passed down as 1500 MTU (technically 1522), and not 1496 (1518)? I would be ecstatic to no longer having to rely on --set-mss 1442 for any VLAN's I pass down from ether1.

I know this MTU bug exists on current V2 systems, including WRAP & x86PC. v3 WAR's work perfectly.

Brad

[tony]

There is no longer MTU issues in regard to VLAN on all v3 releases, including those for the X86-WRAP and the upcoming X86-PC versions.

[mp3turbo]

excellent news, tony, thank you!



only two miracles left:

- superfrequency superfreedom (2312MHz turbo possible? 4900MHz turbo possible? 6000MHz turbo possible? in ## setting)

- bonding for wireless interfaces.



looking forward!

[lonnie]

We cannot enable Turbo on the edge channels like that since it would go outside the assigned spectrum and would really get you in trouble, plus the chip cannot do it since it requires 20 MHz on either side of the channel frequency and the PLL inside the chip is at its limit. It can do 10 MHz but not 20 at the edge.

[mp3turbo]

ok, I understand. I didn't mean to be "totally exact"... if turbo begins at 2322 or 2332, that's ok. if turbo begins at 4920 or 4940, that's ok. If turbo is usable at 2342, 2352, 2362, 2372, 2382, 2392, 2412, 2422, 2432, 2442, 2452 etc, that's totally perfect. In many countries, there are subjects allowed to use for example 2.5GHz spectrum as MMDS is shut down.

you are correct, edge is edge.

[cephlon]

Just wanted to join the Great News Crowd. I can't wait to role this out to my 6 x86 AP's!