Slashdot Mirror

← Back to Domains

Domain: cyclades.com

Stories and comments across the archive that link to cyclades.com.

Comments · 9

  1. Re:Well the top three questions I'd have by zerocool^ · · Score: 1 · on Open-Source Router to Take on Cisco?

    Something that I haven't seen anyone point out yet:

    This has been done for YEARS and YEARS already.

    When I worked at a small webhosting company, we did our own routing. We did it with RedHat + zebra + BGPD + cyclades PC300 cards + Motorolla CSU/DSU's. All the people we bought bandwidth from had no idea what the blind hell we were doing, but they were all amazed at how well it worked.

    • Zebra GNU Routing software (emulates IOS)
    • Quagga (Zebra, but more up to date and maintained by people who accept outside input)
    • BGPD - Supports BGP-4 and multi-homing.
    • Motorola FT100 CSU/DSU (we had the ones with V35's on the back, and custom pinned cables to translate into the 24 pin serial interface on the T1/E1 cards)
    • Cyclades PC300 (we had the PC300/TE-2, I believe, with 2x 24 pin serial ports which required adapters from V.35 out of the back of the CSU/DSU)


    The X-factor here is the Cyclades card. Appearantly, and I didn't know this when I started writing this post, Cyclades (aside from shifting their site to a webserver on someone's AOL 14.4 modem) has stopped selling Wan cards. I would assume someone has taken up the slack, but whatever.

    Anyway, we had 5 T-1's, and this gave us 5 HDLC interfaces, which the linux kernel supports ever since 2.4.something, maybe 2.4.12. We also had our lan port, eth0. I think we had a DMZ on eth1, and then the loopback. Bwm (now bwm-ng) was able to read all the interfaces.

    Interestingly enough, since all of our software was open source, we were able to correct an interesting BGP-related problem. We bought bandwidth from "X" tier-1 bandwidth company, and also from "Y" tier-2 bandwidth company. Company "Y" purchased their bandwidth in large part from company "X", so in essence, we were adding an obfucation step by having the reseller in the picture (but, they offered a lower price, about half). Anyway, BGP is set up to allocate routes to the shortest AS pathlength. That meant that every route that was destined to hit network "X" got routed through network "X"'s T-1 line, and not through the 2 T-1's we had from brand "Y" - essentially we ended up saturating X, and never using Y, due to the way BGP worked. We ended up trying lots of things - prepending our AS paths for the X routes, etc, but eventually one of my coassociates decided to just hack the zebra code.

    I wish we had released it GPL, but the main programmer had said he didn't feel it was well documented enough and certainly used some unclean workarounds (like reading bandwidth stats from ifconfig and not from /proc), so we didn't release it. But, essentially, he found a way to manually allocate routes, by percentage, onto unused T-1's, so we didn't ever get to a situation where we were slamming one T-1, and had 4 un-used. Then, he went on to create a set of sub-routines whereby it would read traffic statistics to properly route traffic based on percentage of bandwidth used versus other links and available overhead. It resulted in a routing system we almost never had to touch, which automatically made the most economical use of our outgoing bandwidth, ensuring that customers always got the most unsaturated link, even if it was longer by one or two hops.

    In the end, we even had the ability via configuration files to add an interface, specify its maximum thruput, and the routing system would automatically take that into account when doing auto route allocation. All because the code was open.

    No, this open routing stuff isn't new.

    ~Will
  2. Re:What does MandrakeSoft gain? by michelcultivo · · Score: 1 · on Mandrakesoft Acquires Conectiva

    No, Marcelo Tosatti is working at Cyclades now.

  3. Re:*drake by tweek · · Score: 1 · on Top Ten Linux Configuration Tools?

    While I agree with your assessment of linuxconf, might I introduce you to my good friend Cyclades?

    I can't seriously imaging not having my ACS32 and PM10s at our datacenter. The only time I've come across a problem that I couldn't fix by using the ACS and PMs was because the RSA proccessor on one of our x445s died and the PM doesn't support the fucked up chinese (electrician's description - not mine) 220 plug on those servers. I've completely b0rked a router after-hours and was able to dial into the ACS, power cycle the router and then console in to redo the config. I used to be downright paranoid of kernel recompiles from afar but with everything from POST to GRUB to OS being redirected over serial, I'd have to REALLY fuck something up to have to drive to the datacenter and fix it.

  4. Re:finally.... by forged · · Score: 4, Insightful · on HP Markets Cheap 4-User PCs To African Schools
    I was still at university in 1996 and the computing science department was running on a SINGLE ibm server with 128Mb of ram, running AIX 4.x.

    This thing ran an Oracle database, mail/web servers, did nfs/nis, supported two dozen X-terminals and at least 50 text terminals. We would run out of memory only very occasionally, when people started doing stupid things like run their window manager on the server itself rather than use the one built-in the X terminals.

    The machine was not fast, but it ran to the department needs and people weren't less intelligent because of this. Maybe quite the opposite in fact: since you didn't want to compile your progamming assignment every couple lines, people were more considerate about writing quality code in the first place and make use of the resources we had more efficiently. This produced a generation of programmers who were concerned about writing good code.

    Today, any a 2GHz+ PC with 1GB+ of ram would put the machine we were using back then to shame in terms of raw computing power and even i/o, with the proper supporting hardware (fast scsi disks, cyclades serial ports board, etc.)

    One last thing about Linux IIRC, part of the memory used by programs is actually shared if two or more users use the same application, so memory utilization is efficient in a multi-user setup.

  5. Re:Been there, done that, painted it metallic gree by stefanb · · Score: 5, Informative · on Rack Mounted PCs for the Home User?
    That being said, a KVM is awfully handy if you need control over a machine during the early boot stages -- you can't get to the BIOS settings display using X forwarding. :-)

    Get a mainboard that supports serial console redirection, like the Tyan Tomcat i875P, and hook the (first) serial port up to a multi-serial board in another box, or get the excellent (but expensive) Cyclades TS-Series console server.

    We just picked up four of them as firewalls (in 1U cases from Chenbro) as well a backup server, and the redirection works like a charm.

  6. Easy to do (If you've got lots of money) by ivan256 · · Score: 5, Informative · on Build Your Own PowerPC?

    If you really want to build a system from parts there are several places you can go. Motorola sells evaluation platforms that consisto of a motherboard (essentially a backplane) and CPU modules that plug into it. It's calld Sandpoint. You can get third party CPU modules for it from Tundra (who also sells whole kits with their own system board). Marvell/Galileo sells a platform that is well suited to building a PC style machine with PPC hardware, and you can get a variety of processor cards for it ranging from low end G3 style processors up to dual 7450 processors.

    Some of the best PPC machines available right now can't be built from parts simply because they're on a single board. My current to y is the cyclades TS-100 it's only 1"x3"x3", has dual CPUs and can be had for under $200.

  7. Good stuff by itwerx · · Score: 2, Interesting · on Experiences Programming on Cyclades Term Servers?

    I've never programmed Cyclades boxes but I've used them off and on over the years as terminal and print servers and I can say they're well nigh indestructible!
    The company's been around forever and their product quality (and tech support) has always been excellent.
    Their website always has useful stuff too.

  8. Re:Cisco's Control by dotgpb · · Score: 1 · on Cisco Eclipses Microsoft As 'Most Valuable Company'
    And to those worried about Cisco's OS, don't worry, Linux isn't going to be challenged anytime soon. The OS is made for administering routers, not for running games and what-not. It's very specific to its task and not exactly something you play around with.

    However, cisco may need to worry about losing market share to linux:
    linux-router.org
    cyclades
    sangoma

    I probably wouldn't use linux as a router everywhere, but it certainly makes sense in some situations. I would guess that competition from linux based routers will cause Cisco to lower the prices on their hardware, at least somewhat.

  9. we use HP LPr and Dell 6350 by the+gaijin · · Score: 1 · on Ask Slashdot: Hardware for Headless Linux Boxes

    i've set up a small farm of linux boxes for serving http, dns, and smtp/pop for employease. i've been using the HP LPr machines: 2U form factor, twin cpu configuration. they have a serial/console out that you can configure linux to use. we are going to send all console interfaces to a cyclades console management system. i'm not sure the LPr's support remote power cycle signals via this interface, but I know the Dell Poweredge 6350's we're using do. unfortunately, these machines currently run NT since Netscape Enterprise server is not yet ported to linux. however, we have it from an insider that Enterprise server will be available in the fall. in the meantime, we are going to test out the quad-xeon-cpu 6350's as linux servers, and will find out if its console interface will actually work with linux.

    PS: Neither of these boxes care if they are headless, handless, or tailless.