Results 1 to 38 of 38

Thread: What do I need?

  1. #1

    What do I need?

    Ok, so months of looking around and talking and trying has gone by. I've tried FreeBSD and while I like it very much, I've decided to ultimately use Solaris.

    Basically here's my deal. I'm NOT a web-hosting company. I'm not planning to become a web-hosting company either.

    I am in fact a college student (well, not much longer) who is looking to start up his business.

    My business specifics I obviously can't go into in depth for obvious reasons. I can say that it will be a public web service that will offer both free and paid services.

    The plan is to make it a largely used resource and to entice users to make it even their home-page. I need to start off with a decent amount of server power so that I have a good starting ground. I plan to start off hosting it with SDSL 1.5/1.5. Basically I need the power to run my site, with the flexibility to add more server power in the future.

    I have decided to use Sun Microsystems servers running Solaris on the Sparc platform. What I cannot decide is what or how many to buy.

    What I had been originally planning was to purchase 3 servers. I was planning to purchase 2 V120's to run DNS, email, and other less powerful services... and then one V210 to run Apache, MySQL, FTP, etc..

    The other option is to run 2 dual processor V210' servers, both running DNS... then have one run MySQL, and one run Apache and email. Or something like that.

    Basically the idea is that I need to run all of my own web-services. My understanding is that I'll need 2 servers and 2 IP's at least because of DNS. My site will be largely php and mysql with some perl basec content to be added in the future.

    My worry is that I don't want to purchase a server, quickly run out of power, and then find that to increase power I need a WHOLE new setup. I want to be flexible. What is the recommendation for a start-up web company/service in terms of equipment.

    My budget is unfortunately not huge, however at the same time it's not set at a particular amount. It's all based on reason.

    Anyways, any help would be appreciated. I have little experience in this realm and it's all quite intimidating. There's so many options and so many variables. As much as I want to just make a decision, I worry that somebody might think of some setup that I haven't thought of yet that makes much more sense.

    Your help would be much appreciated. Thank you.

  2. #2
    Just out of curiousty why did you pick sun microsystems to use?

  3. #3
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083

    Lightbulb

    Here's a bunch of suggestions. Feel free to totally ignore them.

    I'm going to tell you right now, that someone in your position with limited capital budget is far better off with a dedicated hosting environment than purchasing your own equipment and putting it behind an SDSL line. Two dedicated boxes may run you a couple hundred a month for well configured servers, with some administration help, and some bandwidth. That's a lot better than a few thousand up-front. Besides, if things don't pan out, you can cancel your hosting service and they'll just rent those servers to someone else. If you buy servers, you'll have to figure out what to do with them if you fail.

    Second, your decision to go with Solaris on low-end hardware for Apache+MySQL+PHP is, quite frankly, wrong. Sun Solaris has its place, but based on what you've posted, and my read of your experience level, my gut tells me there is no way you should be on Solaris. You'll end up regretting it.

    Third, do run Apache and MySQL on seperate boxes. You'll thank yourself for this down the road. If you go with a dedicated hosting company, make sure they will not bill you for traffic between your own servers. If they say they do, then ask for a private ethernet between your boxes. If they can't do that, well, lots of hosts can.

    Don't administer your own DNS servers unless you've got a ton of zones to publish. It's unnecessary configuration hassle, and it costs you nothing to have your host do it for you.

    Fifth, get more than two IP addresses! I am adamantly opposed to your SDSL thing, but any dedicated hosting place is going to give you more IP addresses if you request and justify them. When you need more, they will provision quickly. The same may not be true with SDSL.

    Finally, since you want to be huge (portal) and you say you'll run PHP/Perl/etc, at some point you'll want to use some kind of load balancers/proxies. Be sure to have a session mechanism that works well with this, if at all possible, without real-server-affinity. Consider using Squid in front of your Apache server now, as it will save on apache child process count / memory consumption, and down the road when you become a huge success, you'll supplement all that with more sophisticated load balancing and HTTP acceleration.

    Right, so, feel free to ignore all of that.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  4. #4
    Join Date
    Jul 2003
    Location
    London UK
    Posts
    327

    Re: What do I need?

    Ok, so months of looking around and talking and trying has gone by. I've tried FreeBSD and while I like it very much, I've decided to ultimately use Solaris.

    Basically here's my deal. I'm NOT a web-hosting company. I'm not planning to become a web-hosting company either.

    I am in fact a college student (well, not much longer) who is looking to start up his business.


    This is all good and fine so long as you are budgeting (time/money) in your business plan for the Solaris learning curve.

    I plan to start off hosting it with SDSL 1.5/1.5. Basically I need the power to run my site, with the flexibility to add more server power in the future.

    I agree with Jeff on this point, SDSL hosting is essentially a bad idea, BUT you will probably get away with it in a testing/development phase, but once you get to a production phase then definitely not a good idea.

    I have decided to use Sun Microsystems servers running Solaris on the Sparc platform. What I cannot decide is what or how many to buy.

    I would again agree with Jeff, start small and rent your servers, you can then swap/grow/upgrade as you need, without having to sell/buy new hardware outright. You can potentially then even go down the route of using FreeBSD because your service provider will be the one having to deal with the hardware compatibility issues, not you.

    Basically the idea is that I need to run all of my own web-services. My understanding is that I'll need 2 servers and 2 IP's at least because of DNS. My site will be largely php and mysql with some perl basec content to be added in the future.

    Don't spend thousands of dollars/pounds on a server to run DNS, unless you really need to host lots of zones and have them under your control, plenty of cheap and good outsourced DNS
    solutions around.

    My worry is that I don't want to purchase a server, quickly run out of power, and then find that to increase power I need a WHOLE new setup. I want to be flexible. What is the recommendation for a start-up web company/service in terms of equipment.

    Again, renting makes more sense, as Jeff quite rightly says, if your project doesn't pan out, you just cancel your rented server contracts without having to deal with offloading kit, less financial risk.

    By all means go down the owned server route once things are picking up, renting sounds like your quickest and easiest route to get started.
    Paul Civati
    Rack Sense Ltd UK Managed Services Provider
    Views expressed are my own and not those of the company.

  5. #5
    the reason I am against renting a server is b/c I've been paying for managed hosting up till now. I've done the math and at this point in time, I've already paid for one server. The way I see it, it's shoveling money down a bottomless pit.

    at least if the business doesn't pan out, I'll be able to get SOME money back.

    As for why Solaris - security/stability/flexibility

    I'm not necessarily looking for the quick, easy, and cheap route here. I'm looking for the right route.

  6. #6
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Originally posted by Inspector Gadget
    the reason I am against renting a server is b/c I've been paying for managed hosting up till now. I've done the math and at this point in time, I've already paid for one server. The way I see it, it's shoveling money down a bottomless pit.
    From your posts so far, I'd say you over-bought server configuration, bandwidth, and so-on. This is your fault, not your host's. Don't buy power you don't need during your development stage. If you can host your site on a 1.5Mb/s SDSL line, you can host it on inexpensive servers. I've been doing what you are doing since 1996, so I have some experience here.

    at least if the business doesn't pan out, I'll be able to get SOME money back.
    Anyone who has done a start-up will tell you two words you should never forget: cash is king. What does that mean? Don't buy something you can lease, rent, borrow, etc. Don't buy capacity you don't need for another 3 months. Don't rent an office that is six times the size you need. Why? In a start-up environment your needs change rapidly, and your cash position makes you flexible, or not.
    If it takes only 3 or 4 months of dedicated server bills to pay for your servers, then so be it. The fact is, if your service grows at all your 1.5Mb/s SDSL won't cut it, and you'll have to deal with shipping those servers to a co-lo, or renting servers anyway.

    As for why Solaris - security/stability/flexibility
    The right reasons to run Solaris are, "Oracle," "Sybase," and uhh.. "Oracle." There are some applications for which Solaris is excellent, but a start-up running Apache+PHP+MySQL ...
    Let's go about it another way. You weren't knowledgable, resourceful, and persistent enough to solve hardware compatibility problems with FreeBSD on an SIS chipset mainboard. What makes you think you will be resourceful enough to solve complex Solaris system administration challenges? Solaris admins make $150k+ even in today's tough job market, and they do so because their job is difficult, requires their years of experience solving Solaris problems, and because as a rule, people who run Sun Solaris have money to blow on it.

    I'm not necessarily looking for the quick, easy, and cheap route here. I'm looking for the right route.
    Take a page from the books of Yahoo and Google.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  7. #7
    Join Date
    Dec 2003
    Location
    Canton, Ohio
    Posts
    62
    If you want my opinion (which I assume your looking for), buy more bandwidth, and less server. I had a Packard Bell 133Mhz "server" ;-) saturate a T1 easily.

    Have you tried to run the x86 versions of Solaris? I am a Linux guy so I have no idea if thats a good idea or not. I know they make a version like that. Might be worth getting a dell / ibm / etc server and installing that. Sounds like your still researching. Maybe you can learn Solaris better in a low-cost startup world before going all out sun servers.

    Dustin Wright
    Network Administrator
    The Karcher Group
    North Canton, OH

  8. #8
    Originally posted by dwright
    If you want my opinion (which I assume your looking for), buy more bandwidth, and less server. I had a Packard Bell 133Mhz "server" ;-) saturate a T1 easily.

    Have you tried to run the x86 versions of Solaris? I am a Linux guy so I have no idea if thats a good idea or not. I know they make a version like that. Might be worth getting a dell / ibm / etc server and installing that. Sounds like your still researching. Maybe you can learn Solaris better in a low-cost startup world before going all out sun servers.

    Dustin Wright
    Network Administrator
    The Karcher Group
    North Canton, OH
    hey... actually it's funny you mention the 133 filling a T1 circuit b/c the author of the book Absolute BSD actually mentions something similar. It's true of course.

    As for the x86, I did try to install it however the install failed every time. However that's not it's fault. It was the same SIS computer that failed with FreeBSD. Unfortunately I don't have a spare computer at the moment to try it on. What I have heard from just about everybody who is familiar with it is that it's not quite ready. Apparently Solaris runs ridiculously slowly on x86. Everybody has recommended that I go with Sparc if I decide to use Solaris. I don't quite understand though why everyone keeps saying that it's SO much more expensive. I've been looking at the prices and while it is more expensive than a generic box would be, it's not that bad.

    Now on the other hand, my reasons for using Sun/Solaris could also be fulfilled with an HP-UX rig or an AIX rig.. after looking into them though it seems that they're even more expensive than Sun systems. Am I correct? I want to go with a commercial hardware/OS solution for stability and power however b/c I'm not currently using anything that requires Solaris itself, my needs could also be met most likely with other commercial solutions. I just don't want to build another FreeBSD box and worry about drivers etc.

    Also, JSW6, before you act like an ***, at least know all the facts. It's not that I wasn't resourceful enough to resolve issues with my FreeBSD box, it's that I just didn't care enough at the time to bother with it. If there's one thing that I can't stand, it's someone trying to tell me, "You're not smart enough, which is why you should hire me." - which is what it seems you're trying to say.

  9. #9
    also while I thought it was implied, the plan is that I will start with SDSL 1.5/1.5, and if my needs exceed that, I'll colo it. I see no need to colo it at the moment while it starts up. People need to discover it before it starts getting heavy usage. That's why I want to concentrate on the server first. Bandwidth is an easy issue. If the server doesn't cut it though, that's more work and money.

  10. #10
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    I don't need a job from you. I'm giving you what you asked for in your post, the benefit of my experience. There is no sense in the combination of a start-up web portal, SDSL transit, and thousands of dollars in Sun hardware. I wish you success, but I doubt that, a few months from now, you will be anything but disgusted with the amount of time you invest in the Solaris learning-curve, and in making your applications work correctly in a Solaris environment.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  11. #11
    well, since it is agreed that mySQL and Apache should be on separate servers, would 2 SunFire V210's cut it? Can one machine run BIND/mySQL and the other BIND/Apache ? Or must it be a separate machine? If that would work, it would take care of everything pretty much.

    When I say the V210, I'm talking the large configuration:

    Sun Fire V210 Server
    2 1-GHz UltraSPARC IIIi Cu Processors
    1-MB L2 Cache per Processor
    2-GB Memory
    2 36-GB 10000 RPM Ultra 3 SCSI LVD Disk Drives
    4 10/100/1000-Mbps Ethernet Ports
    1 66-MHz, 64-bit Wide PCI Slot System Configuration Card
    Sun Advanced Lights Out Manager (ALOM) Pre-Installed
    Solaris 8 Operating System Pre-Installed

    Or I could go with the small, but I don't know how adequate it would be:

    Sun Fire V210 Server
    1 1-GHz UltraSPARC IIIi Cu Processor
    1-MB L2 Cache per Processor
    512-MB Memory
    1 36-GB 10000 RPM Ultra 3 SCSI LVD Disk Drive
    4 10/100/1000-Mbps Ethernet Ports
    1 66-MHz, 64-bit Wide PCI Slot System Configuration Card
    Sun Advanced Lights Out Manager (ALOM) Pre-Installed
    Solaris 8 Operating System Pre-Installed

    The other option is to start with the small and then expand where necessary as time goes on.

    Still though I have limited storage either way so I'd want to get some type of SATA NAS system I'd assume.

    Let's start this topic over, for the moment accepting and assuming that I'm going to go with Solaris. Rather than talking about admin issues or whatnot, let's talk from a hardware standpoint purely right now. The hardware is the starting point. From there I can work on bandwidth. I'm stuck on all aspects of hardware though as I am not planning to use any of my current equipment as it's no more than home/consumer grade equipment.

  12. #12
    Join Date
    May 2002
    Location
    Sunny California
    Posts
    1,679
    Don't use Sun equipment. It's not the right tool for the job.

    This is coming from someone (yes, me!) who:

    a) Worked at Sun for more than a year;
    b) Converted some of Sun's intranet servers to Linux;
    c) Has a Solaris admin certification.

    Read my lips: Solaris isn't the right answer as an Apache/PHP/MySQL web development platform. Even Sun will tell you that. That's why I embarked on a (successful) project while I was there to convert a bunch of Sun's internal web servers to Linux. Everyone wanted to be able to use PHP and MySQL. PHP and MySQL are a b*tch on Solaris (Apache is not as much so.)

    If you wish to pay for software, start out with RHEL 3.0. If not, run Red Hat 9 or Debian. Grab a control panel like DirectAdmin for your web server; that'll set you up well. If you don't want to use a control panel, set it all up by hand, but setting it up by hand is not worth much except as a learning experience at this point.

    Get a real colo solution. DSL will not be as stable as you want it to be, and the migration will be a pain in the butt once you want to scale past 1Mbit.

    Get two 1U servers and use one for MySQL and one for Apache/PHP. This is the right thing to do for scalability.

    Listen to Jeff. He knows what he's taking about (based on the posts of his I've read in this thread.)

    If you need specific help, feel free to PM me.
    Erica Douglass, Founder, Simpli Hosting, Inc.
    I founded Simpli Hosting, and sold it in 2007 to Silicon Valley Web Hosting after over 6 years in the business.
    Now I'm blogging at erica.biz!

  13. #13
    Within this community I've had a large number of opinions that Solaris is a bad server.

    I'm not opposed to hearing negative things about Solaris as I don't totally have my mind made up yet. However I really would like a bit more detail as to why.

    You say that mySQL and PH{ are a "b*tch" on Solaris. Why is this? What has gone so terribly wrong in Solaris that you say this? Again, I'm not opposed to hearing negative comments, but please tell me more as I'm entirely interested.

    the DSL bit I'm pretty set on though. I'll deal with the migration when the time comes, but as was said, I shouldn't pay for resources I won't be using, and I won't need serious bandwidth for at least a year which would waste money if I colo'd. When I reach the point where I need more bandwidth, I'll colo.

    Also you agree with my using two servers, one for mySQL and one for PHP/Apache. Can I also run BIND/DNS off of these two servers or must they be 2 separate machines? I actually am going to have many domains, however all fowarding to specific folders on the same website, so I do need to run my own DNS. If I could run it all from two servers, that would really be optimal.

    Also you mention mainly Linux, I'm not really interested in touching Linux at the moment until the smoke clears.

    Please, I'm interested in hearing WHY Sun/Solaris is a bad server b/c I've heard it several times now on this forum. I just really need more input than people telling me "what not to get". I'd be more receptive if I got "what not to get" - "why not" - "what's a better option" - "why it is". At then it seems to be less opinion and more fact.. you know? It just makes more sense to me that way.

  14. #14
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Solaris isn't "not a good server," so much as it is not a good platform in the hands of an inexperienced administrator. If you had years of Linux, FreeBSD, etc. experience you'd still have a huge learning curve to tackle with a Solaris installation. That's the single biggest disadvantage Solaris has versus other unix platforms for someone like you (or me, or anyone else who is not a Solaris expert.)

    What platforms are better for your use? Well, I'll tell you that Linux is the best platform to run MySQL on, even though you want to stay away from it. MySQL is developed for Linux, and it has excellent SMP performance on Linux, where as FreeBSD, for example, cannot execute multiple threads concurrently until 5.x (?), which is a signifigant disadvantage.

    That said, I think you'd be happy with FreeBSD if you are unwilling to "risk" Linux.

    As to your BIND question, you should have no trouble in the configuration you suggest. While I still think an outsourced or ISP supplied authoratitive DNS server will better allow you to focus on your web application, it shouldn't give you problems.

    Also, I'd really take a hard look at the amount you spend on server hardware while in your development stage. You mentioned that it'll be a year before you have enough traffic to exceed what your 1.5Mb/s SDSL line can serve. You can figure how many pageviews/day that will support based on the size of the graphics and HTML you generate. Your connection will limit the number of clients who can access your site, and thus, you may not need such powerful hardware.

    The Sun boxes you spec'd were pretty powerful for 1.5Mb/s worth of web serving, and also, pretty expensive for a one man start-up. (Sorry if I'm off the mark there, you mentioned you would not employ a Solaris administrator, etc.) I urge you to conserve capital and purchase servers that will just be adaquate to get you through your development cycle, not adaquate for your real deployment in a co-lo/etc.

    What I'm trying to say is, you won't benefit from 3GHz Xeon servers with fast SCSI disks when used 1GHz Celerons with ATA will get the job done easily, and keep cash in your war chest. As you gain experience with your operation, MySQL, Apache (+Squid hopefully!), and so on, you will have a solid idea of what your bottlenecks are and what kind of configuration you need for your "launch."

    Finally, you mentioned NAS boxes a couple of times. I suggest you do a ton of research and talk to several people who have used each NAS platform you are considering before you make any kind of purchase. Consider both your storage size needs, and the application's performance needs, when evaluating units. For example, MySQL is not so great over NFS; however Linux (example) handles accessing multiple filesystems better when they are NFS and not locally attached ATA/SCSI devices.

    The best way to spend a lot of money on something you hate is to buy a huge EMC / A10000 / Shark installation and later find out that MySQL (for example) didn't benefit from it the way you hoped, or that you bought 600GB of capacity to get you through the next 18 months, but 8 months later the cost of disk shelves went down 30%!
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  15. #15
    My suggestion would be to hit ebay and try and get 4 odd V100's for about $500 each, and 1 bigger machine for MySQL.
    Use Squid as a front end proxy on one box, run all the web services on the others. Clustering is king, and with more boxes you can easily pull one out of production for development etc.

    I agree with the other posts - a single box will be enough to saturate your SDSL, but assuming you want to grow smoothly, starting with a cluster will make scaling it up easy.

  16. #16
    Join Date
    Dec 2003
    Location
    Canton, Ohio
    Posts
    62

  17. #17
    true... I was looking to go aftermarket for Cisco stuff actually.

    so let me get this straight:

    You guys are saying that Sun is a good server, but only in the hands of a good admin? I'm trying to get this straight. Some people are saying that Sun is a bad server, some are saying it's a great server, some are saying it's a good server, but only on Mondays and Wednesdays, and some are saying it's a good server only in the hands of a good admin. What's the bottom line about Sun? Good server? Bad server? Should I for some reason be looking at HP or IBM instead?

    also, what's the deal with Squid? How does that work? How does that work into my config?

    I agree with starting out with something that can be expanded into a cluster. However I'm unsure how to go about doing that. Would I be able to use V210's to build a cluster, or do I need another type of system like a blade system to do that? As you guys have figured out, I am somewhat of a newbie, however on the other hand, being young, I have all the time in the world to learn. This is why the idea of a learning curve doesn't scare me. I don't mind learning to use something if it's the RIGHT thing to use.

  18. #18
    Join Date
    Feb 2004
    Posts
    634
    I like Sun hardware--it's been very reliable for us and their support is topnotch, and they are definitely getting more competitive on the pricing front. Unfortunately I think this may be a bad time to buy low-end Sun servers. It's pretty obvious based on the marketing blitz they've unleashed in the last month that they are heavily promoting Opteron on the low end; it seems like every tech-oriented magazine and large website is running ads for Sun's soon-to-ship V20z Opteron server. The price/performance ratio of Opteron just kills their existing low-end offerings, especially those based on the Ultrasparc IIIi "Jalapeno" core like the V210. The higher end boxes currently based on the Ultrasparc III will supposedly be able to upgrade to the "multiple core on a single chip" design of the Ultrasparc IV, so I'm not sure where that leaves the IIIi on the roadmap.

    My company is an iForce member, and we've gotten to play around with Solaris 10 and the V20z. According to my staff, this will be a very compelling offering for low end web work, so you may want to wait a bit longer if you can. It should be shipping very shortly (if it isn't already); I believe initially it will run a 32-bit version of Solaris 9 for x86, with a 64-bit version to follow shortly. Solaris on x86 used to be regarded as the ugly stepchild in the family, but with their focus on Opteron on the low end I expect Sun to put some significant resources into making it better. In our experience, Solaris 9 on x86 isn't nearly as bad as those here have made it out to be. If you use it with Sun hardware, it is much easier to deal with as it has pretty marginal third party device driver support compared to Linux. Solaris 10, which is due out later this year, looks to be a very good product from our beta testing. From what I've read, performance in Solaris 10 should be better for lower end systems. Previous versions of Solaris were heavily tuned for multi-processor, multi-threaded usage, which led to performance degradation due to wasted overhead and locking on uni-processor, low thread environments. But they have apparently addressed that, and there are some other cool features coming in Solaris 10 like predictive self healing, better process rights management, diagnostic tracing, etc.

    I don't understand the belief that Solaris is somehow much more difficult to grasp than any other Unix. My company primarily develops hosted applications in Java, and we have deployed things in a variety of environments. Our sysadmins that are more comfortable with Linux pickup what they need to know about Solaris with the same relative ease (or difficulty) as those with Solaris knowledge pickup Linux. Solaris sysadmins do make more money, but I don't think that's inherently due to any additional complexity of Solaris but the fact that many enterprises currently deploy Linux to less critical edge applications and Solaris to more critical core apps. I am far from a certified sysadmin--I'm in technical management, meaning I know enough to dangerous but not enough to actually do much :-)--and as a weekend project I picked up a Sunfire V100 and set it up as a web/app server running our low-end Java web setup (combination of Resin app server and PointBase database) off my DSL line at home. It wasn't too terribly difficult, and there are several guides available online to help setup a Solaris server securely for web services, FTP, email, etc. The biggest problem I encountered was in compiling certain things, as Sun's compiler doesn't always work with things expecting the Gnu compiler. But if you install the Solaris gcc complier you should be fine. The Solaris package includes a CD-ROM containing many of the binaries a sysadmin would need, and as has been mentioned the sunfreeware.com site has compiled Solaris binaries (Sparc and x86) for most of the popular open source software .

  19. #19
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    You won't have Squid worries until after your "launch" for sure, but when you get there, the purpose of Squid in a serving environment is to make more effective use of your Apache process count, and thus memory utilization.

    Let's say you have MaxClients 100, and each of your Apache+PHP+KitchenSink processes uses around 3MB of non-shared memory. Not bad, but when you get popular and MaxClients looks more like 1000 than 100, 3MB/process is really valuable memory. Eventually you become memory-bound, as opposed to I/O or CPU -bound, so you drop more money on ram, right? Neah.

    Most of the time, those 100 (or 1000) processes are either waiting for a request from the HTTP client, or transmitting same request to the client. They may do 15ms of work for a client that is connected for several seconds. Why? HTTP clients are slow. They are 56Kb/s modem users, or perhaps 1.5Mb/s DSL users.

    Squid, on the other hand, would be connected to your ethernet and configured as a reverse proxy, meaning HTTP clients connect to Squid, and Squid sends the request on to your web server. Note that it is not a cache in this configuration.

    The advantage is, Squid rapidly sends the request to your Apache process, and receives the response, at several mbit/sec (depending on client/network load.) There is also virtually no latency, unlike a modem or DSL user who may be a few dozen milliseconds or more away. The TCP handshake happens faster, the request happens faster, the computations for PHP and whatnot still take just as long, but the response is downloaded from Apache by Squid faster, and then that Apache process is free to handle another request.

    This reduces your client count per the same number of simultaniously connected users (they are just moved to the Squid front-end) and also cuts into context swap penalty, meaning that going from 1000 processes each doing 15ms of work every 3 or 4 seconds to 250 processes doing 15ms of work every 1 second will slightly increase overall system efficiency. That's just a side-effect though, and not the goal.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  20. #20
    Originally posted by jsw6
    You won't have Squid worries until after your "launch" for sure, but when you get there, the purpose of Squid in a serving environment is to make more effective use of your Apache process count, and thus memory utilization.

    Let's say you have MaxClients 100, and each of your Apache+PHP+KitchenSink processes uses around 3MB of non-shared memory. Not bad, but when you get popular and MaxClients looks more like 1000 than 100, 3MB/process is really valuable memory. Eventually you become memory-bound, as opposed to I/O or CPU -bound, so you drop more money on ram, right? Neah.

    Most of the time, those 100 (or 1000) processes are either waiting for a request from the HTTP client, or transmitting same request to the client. They may do 15ms of work for a client that is connected for several seconds. Why? HTTP clients are slow. They are 56Kb/s modem users, or perhaps 1.5Mb/s DSL users.

    Squid, on the other hand, would be connected to your ethernet and configured as a reverse proxy, meaning HTTP clients connect to Squid, and Squid sends the request on to your web server. Note that it is not a cache in this configuration.

    The advantage is, Squid rapidly sends the request to your Apache process, and receives the response, at several mbit/sec (depending on client/network load.) There is also virtually no latency, unlike a modem or DSL user who may be a few dozen milliseconds or more away. The TCP handshake happens faster, the request happens faster, the computations for PHP and whatnot still take just as long, but the response is downloaded from Apache by Squid faster, and then that Apache process is free to handle another request.

    This reduces your client count per the same number of simultaniously connected users (they are just moved to the Squid front-end) and also cuts into context swap penalty, meaning that going from 1000 processes each doing 15ms of work every 3 or 4 seconds to 250 processes doing 15ms of work every 1 second will slightly increase overall system efficiency. That's just a side-effect though, and not the goal.
    Ok, so then I assume that SQUID must be on a separate machine as Apache? Or would it be logical to place Apache on a port other than 80 and then put SQUID on port 80 directing to another port? Although I would guess that this would defeat the purpose of SQUID in the first place no?

  21. #21
    Originally posted by lockbull
    I like Sun hardware--it's been very reliable for us and their support is topnotch, and they are definitely getting more competitive on the pricing front. Unfortunately I think this may be a bad time to buy low-end Sun servers. It's pretty obvious based on the marketing blitz they've unleashed in the last month that they are heavily promoting Opteron on the low end; it seems like every tech-oriented magazine and large website is running ads for Sun's soon-to-ship V20z Opteron server. The price/performance ratio of Opteron just kills their existing low-end offerings, especially those based on the Ultrasparc IIIi "Jalapeno" core like the V210. The higher end boxes currently based on the Ultrasparc III will supposedly be able to upgrade to the "multiple core on a single chip" design of the Ultrasparc IV, so I'm not sure where that leaves the IIIi on the roadmap.

    My company is an iForce member, and we've gotten to play around with Solaris 10 and the V20z. According to my staff, this will be a very compelling offering for low end web work, so you may want to wait a bit longer if you can. It should be shipping very shortly (if it isn't already); I believe initially it will run a 32-bit version of Solaris 9 for x86, with a 64-bit version to follow shortly. Solaris on x86 used to be regarded as the ugly stepchild in the family, but with their focus on Opteron on the low end I expect Sun to put some significant resources into making it better. In our experience, Solaris 9 on x86 isn't nearly as bad as those here have made it out to be. If you use it with Sun hardware, it is much easier to deal with as it has pretty marginal third party device driver support compared to Linux. Solaris 10, which is due out later this year, looks to be a very good product from our beta testing. From what I've read, performance in Solaris 10 should be better for lower end systems. Previous versions of Solaris were heavily tuned for multi-processor, multi-threaded usage, which led to performance degradation due to wasted overhead and locking on uni-processor, low thread environments. But they have apparently addressed that, and there are some other cool features coming in Solaris 10 like predictive self healing, better process rights management, diagnostic tracing, etc.

    I don't understand the belief that Solaris is somehow much more difficult to grasp than any other Unix. My company primarily develops hosted applications in Java, and we have deployed things in a variety of environments. Our sysadmins that are more comfortable with Linux pickup what they need to know about Solaris with the same relative ease (or difficulty) as those with Solaris knowledge pickup Linux. Solaris sysadmins do make more money, but I don't think that's inherently due to any additional complexity of Solaris but the fact that many enterprises currently deploy Linux to less critical edge applications and Solaris to more critical core apps. I am far from a certified sysadmin--I'm in technical management, meaning I know enough to dangerous but not enough to actually do much :-)--and as a weekend project I picked up a Sunfire V100 and set it up as a web/app server running our low-end Java web setup (combination of Resin app server and PointBase database) off my DSL line at home. It wasn't too terribly difficult, and there are several guides available online to help setup a Solaris server securely for web services, FTP, email, etc. The biggest problem I encountered was in compiling certain things, as Sun's compiler doesn't always work with things expecting the Gnu compiler. But if you install the Solaris gcc complier you should be fine. The Solaris package includes a CD-ROM containing many of the binaries a sysadmin would need, and as has been mentioned the sunfreeware.com site has compiled Solaris binaries (Sparc and x86) for most of the popular open source software .
    that's ok.. I can actually stand to wait a bit. I am afterall still researching this anyways. I wasn't planning to dive in immediately. I have probably 2 months or so I can wait before actually making any purchases. I've seen the ads for the V20z. If it turns out to be a good choice, I'll have no problem going that route. I hope that Solaris 9 or 10 really fixes the x86 performance issue as that was my primary reason for looking at Sparc over x86. Everyone has been just telling me not to bother with it in the current state at least.

  22. #22
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Originally posted by Inspector Gadget
    Ok, so then I assume that SQUID must be on a separate machine as Apache?
    By the time you are large enough to need this, you will want to seperate Squid from Apache. Your Squid box may be the perfect place for utility services like DNS, etc. though.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  23. #23
    ok, so after thinking about this, I think the optimal setup would be actually 4 servers.

    2 smaller servers - #1= DNS/SQUID #2=DNS/EMAIL

    2 larger servers - #3=Apache/PHP/Perl/FTP #4=MySQL

    After looking at Sun servers some more (particularly the warranty) I can understand the statements of them being somewhat expensive.

    The next question is, how are IBM AIX servers and how are HP HP-UX servers? Quality/Price? It may be better for me to go with another option so that I can get more for my money to start off.

  24. #24
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Originally posted by Inspector Gadget
    The next question is, how are IBM AIX servers and how are HP HP-UX servers? Quality/Price? It may be better for me to go with another option so that I can get more for my money to start off.
    I guess we're going through this again. AIX and HP-UX are both bad platforms for what you want to do. To be blunt, you need to get off your commercial unix horse and run Linux or FreeBSD. You'd be better off with Solaris than HP-UX or AIX, but the fact is Linux is the best platform for your job, probably followed by FreeBSD.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  25. #25
    http://uptime.netcraft.com/up/graph/...whitehouse.gov

    the president can't be wrong can he? jk

  26. #26
    fair answer jsw6. OS aside for a moment though, what about that configuration I listed. I was assuming that SQUID doesn't need a powerful machine. If I'm wrong though let me know. What should be the specs for a machine running SQUID? That is assuming that it IS caching.

  27. #27
    also I have a conference call with a reseller for Sun on Monday. If I am not happy ultimately with the idea of using Sun, then I'll revisit Linux/FreeBSD. Although I will admit I'm somewhat shaky about going that route simply for hardware reasons. Just because the hardware is listed as being "supported" doesn't mean jack squat in my brief but revealing experience.

  28. #28
    oh and also I should apologize for becoming heated earlier in this thread with you jsw6 as you have been quite helpful. I misread/misunderstood where you were originally going with this whole thing and should not have said what was said. I do appreciate your help very much.

  29. #29
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    You won't be using Squid in a caching configuration, as the purpose of it is to free up httpd processes more rapidly, not to off-load content delivery. If you want to host the images and other non-dynamic content on a platform other than Apache+PHP (which is smart) you should either out-source it to someone who does that well, or run thttpd for your images. thttpd is very light-weight, and will perform better than Squid as a static content delivery server.

    I'd buy roughly the same configuration for your web serving cluster and your Squid box/boxes. You should still use the right configuration for the job, but if you can condense your configurations to two and use one of those for every role, you will have an easier time managing your computing resources. It'll be easier to repurpose that under-utilized Squid server as another web cluster node when traffic goes up, or pull a node out of the cluster to act as a mail gateway when you want to run that new anti-virus scanning package on your users' incoming mail. Flexibility is key

    As far as hardware support goes, just buy your Linux or FreeBSD servers from a vendor who has experience with those platforms. I like www.fnordsystems.com, although you may be more comfortable with a vendor like Dell or IBM, who also sell good, expensive servers with support arrangements.

    Frankly, the hardware compatibility problems you experienced with Linux(?) and an SiS chipset mainboard are nothing compared to the OS headaches you will encounter with HP-UX.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  30. #30
    IBM sells FreeBSD systems? I did not see this on the site however this would be perfect. Where can I find out more about that?

  31. #31
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    IBM sells Linux systems. I think Fnord Systems will sell servers with FreeBSD installed. You really do need to make a choice between a FreeBSD environment without vendor support (not so bad), a Linux environment with support from IBM/etc (good but expensive), or a Solaris environment (which is totally wrong for your needs.)

    If I were you, I wouldn't put a lot of stock in the SCO vs IBM lawsuit. Even the judge is growing impatient with SCO's tactics; SCO has been barred from making many of the statements they made in their accusations against IBM in any form in Germany, as they were unable to prove those statements to be factual to German courts' satisfaction.

    Also, Mr McBryde is encouraging the federal government to ... do something, he hasn't decided what yet ... because Linux is a weapon and because rogue countries such as North Korea (his example, here) can use it to create super computers.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  32. #32
    Join Date
    Jan 2003
    Location
    Chicago, IL
    Posts
    6,889
    You can use Playstation 2's to create supercomputers as well, I say we ban the manufacturing of them as well....
    Karl Zimmerman - Steadfast: Managed Dedicated Servers and Premium Colocation
    karl @ steadfast.net - Sales/Support: 312-602-2689
    Cloud Hosting, Managed Dedicated Servers, Chicago Colocation, and New Jersey Colocation
    Now Open in New Jersey! - Contact us for New Jersey colocation or dedicated servers

  33. #33
    Originally posted by KarlZimmer
    You can use Playstation 2's to create supercomputers as well, I say we ban the manufacturing of them as well....
    I agree! I've always thought that they were dangerous.

    To be honest, you could probably use the electronics in Tickle-Me-Elmo's in a dangerous way if you had enough of them. Is that what we want to give our children? Weapons?!



    back to topic for a sec.... despite my agreeing with the majority that SCO's claims won't ultimately hold, I still think that it's a bad time to get involved personally. While the smoke WILL clear, it's just a matter of time... until then though, I'll go Unix. Plus I've been devoting my time to learning Unix system administration, so I feel like it'd be best to at least use Unix for a while and take a break before I attempt to transition over to Linux. Although my small experience with Linux has led me to believe that transitioning won't be a big deal.

  34. #34
    my biggest hurdle I guess is really whether I want to continue on my quest to use a commercial Unix, or whether I should just go with FreeBSD again.

    While for support reasons I'd like to go commercial, I can see the money that I'd save by using FreeBSD and less costly (non-name brand) hardware. It really comes down to performance and stability and whether it would be just as stable. While I might run into hardware compatibility issues, the bottom line is that with the money I'd save, I can actually afford to replace hardware if need be... but the flipside is whether I want to take the time to deal with that.

    Again, I'll wait until my meeting/conference-call tomorrow and see what happens.

  35. #35
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    It doesn't come down to performance and stability. The problem here is you've got some huge misconceptions about the value of commercial unixes, and you can't let go of them. Do you think people still buy new HP-UX installations for new projects in start-up environments? "No." HP-UX is sold in places where it has been entrenched for years.

    Is Solaris sold in start-up environments? Of course it is. Solaris has huge advantages over other platforms for some tasks. Solaris has OS features which are specific to making it run Oracle better. Solaris on Sun big iron can survive online CPU replacements. I don't think Linux will be doing that anytime soon. However, you will not be swapping out failed hardware on your running web servers anytime soon.

    FreeBSD, and Linux, are fine platforms, and are not inherently less stable than Solaris, HP-UX, or AIX, for the task at hand.

    We've been back and forth on this thread quite a lot, and I really think it's time for you to forget everything you believe about the present value of commercial unices. It's no longer there for web serving. It's been gone for years, and you seem to be living in the past.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  36. #36
    Is Solaris really considered a commercial Unix by itself? I thought that the deal with it was that the license to use it was free but the media wasn't?

  37. #37
    Join Date
    Oct 2003
    Location
    Atlanta
    Posts
    63
    Just a reccommendation - If this project is so important to you why do you "Rent" the Sun Servers for 1-3 months and do your testing on them. You can get them from www.atlantixglobal.com

    I know a girl thier Page Dispenza 770-582-7222 giver her a call and run your questions by here. "She is a Sun Guru" it's worth a try.

  38. #38
    thanks mreedy.

    actually after my conference call today, I'm beginning to understand what people are talking about. I'm starting to feel hesitant about Sun servers and Solaris.

    so, now I guess I start over again. Problem is, I hate Dell. So how's IBM for FreeBSD systems? I don't mind going with FreeBSD as my Unix. As long as it's Unix and not Linux, I'm happy.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •