• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Designing webhost from the ground up

TurboFC3S

Junior Member
I'm going to be building and configuring a webhost from the ground up. We don't have much money to burn, but getting the best performance out of what we can spend is the goal.

My ideas right now are to build a pair of DNS servers. I have a couple extra PII 400 CPU's and 256mb ECC RAM modules, so I'll use those. Get a Supermicro i810 slot 1 mobo, rackmount case, and 3Com NIC's. I can build those machines for about $500 each. Probably use either Win NT 4.0 or FreeBSD for the OS. I already have a ton of NT liscenses.

That's the easy part, the rest is tricky. I can't decide if I'm better off putting money into a big SCSI disc array, or tons of RAM for the webservers and a tape backup. Right now I'm thinking a pair of load-balanced Athlon 1ghz webservers with IDE drives, and a database/Cold Fusion server with a RAID 5 array. Anyone know how the built-in loadbalancing of Win 2K Advanced Server works? Is it a CPU hog? I would put about 512mb RAM in the webserver and 1gig in the DB server. Back up all the webserver data to the RAID array for safety. Actually I would probably use the internal IDE drive of the webserver for just the operating system, and writing log files. For the database machine, I'm thinking about getting one of these discontinued HP Netserver's I'm seeing all over the place like on www.ubid.com There's some good deals on them right now. Anyone know how they perform? Like this thing - http://www.ubid.com/actn/opn/4053243.asp

Another question is what to do for a cost-effective firewall. I'm clueless on this, so can anyone offer suggestions? Am I better off using a box with a normal OS and firewall software on it, or buying a firewall appliance? I also have a Cisco 3524 switch, so that's not a problem. Get a rack, KVM switch, and battery UPS and I should be set. I already have all the necessary software like Win 2K, Cold Fusion Server, SQL 2000, Veritas Backup Exec...etc.

What other things should I think of? And where is a good place to go online for primers on setting up your network security? Like which ports to close, and how to do auditing for intruder detection? I'm a rookie at that kind of stuff.

This is my first post on this forum, sorry for rambling...but I hope to get some answers. And hopefully provide a few answers along the way as well. Thanks.
 
Wow, you're really going all out. The only comment I have is about the SCSI storage. I don't think it's really necessary for you to spend a ton of money on SCSI when IDE is very comparable in performace yet cost much less. You could spend the money better on a lot of RAM and backup measures like mirroring or tape backups.

What kind of connection to the Internet are you going to be hosting your webhosts on?
 
IDE faster than SCSI. You're smoking crack.

Database servers = tons of RAM, SCSI disk-striped/mirrored (hopefully fiber channel) high cache CPU (a.k.a no AMDs) intel or SUN.
Front end servers = tons of RAM, disk not so important, high CPU (AMD ok here)
3524 switch = plenty for small network. Can use VLANs for separate networks between internal/external/DMZ
Firewall = cisco pix or the like, model depends on bandwidth requirements and processor reqs.

Database servers are most often bound by I/O. Meaning the physical speed of the disk or the number of requests hitting a single disk are the bottlenecks. If you want your database to humm then 1) make sure the database design is as efficient as possible. 2) make sure the disk subsystem takes advantage of all technology possible at the time. Always log to a separate disk/controller.

My .02 I'll probably get into trouble for preaching but that's the way I see it.

cheers!
 
Dude, I didn't say IDE was faster the SCSI, just that the performance was comparable. Chill. Unless he's going for a bunch of 10k or 15k scsi's then the peformance will be comparable. If he did go for a bunch of 10k and 15k HD's then he's going to spend a ton of $$$.
 
I would say SCSI due to multitask.

Imagine having more than 10 people writing to the disk through your FTP access and 100 others requesting data from the disk (webserver). SCSI helps improve performance in that aspect.

IDE can be sometime faster than SCSI, that's true.. however in a multitask server environment SCSI smoke asss.

For a starter, IDE environment is fine. You can always upgrade later.


How many servers are you planning to build ? You seem to know what you're talking about. Thats good. What are you gonna use for your webhost management tools? rather than spending tons of $$$ into hardware, you might as well look into the issue of server management tools for your customers

eRr
 
Well, it's not going to be your traditional webhost. I have a couple of websites right now that generate around 10 million impressions/month with most of the traffic coming from an Ultimate Bulletin Board forum. My main goal is to build a home where these sites can grow and have plenty of resources in the future. However, I plan to also host other sites to help cover the bills. I'm fortunate in that most of the sites I will host will be those of friends...all word of mouth. I think our needs will be very much like those of this Anandtech site. We will have the static content, and the forums.

I think right now I'm just going to build one hoss of a server and put the static and dynamic content on it. There will still be the 2 DNS servers and separate e-mail server. Then in the future add a couple webservers, like AMD machines with IDE drives. But get one machine that can do it all now, and in the future can be a dedicated database server. Something with dual channel RAID, dual high-end NIC's, 10K RPM drives, quad-CPU capability, and at least a gig of RAM. Is fiber channel that much faster? I don't really know anything about it. What other tricks are there that I can use?

I'm also still curious as to options for a cost-effective firewall, and for where I can get some tips on network security. Anyone know of good websites that show you what all to secure?

Thanks again...
 
Another question, I've got this program... BlackIce Defender, but I don't know what it does exactly. Somebody told me that it can handle all my security needs.
 
blackice is an entry level firewall program. It might work ok for a small webhost. However for a better security, I would recommend a hardware based firewall like cisco pix, sonic guard or F5 labs big ip.

I don't think CPU matters that much with static content sites. something like p3 700 can go a long way.

eRr
 
Major overkill for boxen just doing DNS as well. Trust me DNS doesn't require much. We just replaced our primary DNS server with about what you're considering and it was a PPro 166 w/256MB of RAM and that acted as both authoritative 4,000+ domains AND resolver for 120,000+ subscribers.

Since you've got 'em laying around though...what the heck. Should be good and fast.

As to disks, spidey has put it well here. I might add that if this is production stuff SCSI is MUCH, MUCH, MUCH more reliable that IDE. It's commercial versus consumer products here. Under heavy load the SCSI system will outperform the IDE system by leaps and bounds and keep humming steadily away for years.
 
You were asking about the firewall...If you're gonna do this thing right, I would suggest getting a good Cisco router...I've not kept up with their best models because I used to work for a University and they were all pretty much just thrown in the basement and given to the primary server guy down there (scary guy...didn't get out much and looked like he was a zombie or something). But you will want to have a way to shutdown all the ports you do not use to avoid the hackers and crackers...
 
I know the dual DNS servers are overkill, but those need to be redundant...and I can do them cheap, so what the heck. I know SCSI is important for that stuff that has many simulatneous read/write requests, all the IDE I/O is handled by the CPU, but SCSI has it's own subsystem. I understand the importance of SCSI. I'm still strugling with what to do for a firewall. I like the idea of something like a Cisco 2600 series router/firewall, but those aren't cheap. Having a software firewall on a box I can build is more attractive for me, but chosing the right software is something I don't know about.

What would you think about having a pair or Athlon 1ghz, IDE machines for the static web frontend (similar to what anand did in their server upgrade). But put high speed SCSI cards in the webservers, and tie them to an external SCSI RAID array? Maybe a fiber channel array. What would be the benefit of having an external array instead of a server with an array built in?

Thanks again for your help guys!
 
I won't touch the firewall question as I only dabble there.

As to why an external array? One reason is heat. Another reason is that you can get more drives in an external array setup that you can inside an single case. We provide usenet for 500,000 subs on a system that is comprised of 73GB Seagate Cheetahs...4TB worth of news in RAID 5. That's 4TB storage. That's A LOT of drives!

Oh, btw, this setup currently stands at 160 73GB drives.
 


<< around 10 million impressions/month with most of the traffic coming from an Ultimate Bulletin Board forum >>



I'd just point out for anyone who hasn't dealt with it, that UBB is a pig. It REALLY stresses the I/O subsystem, and sucks up storage like Rosie does danish pastries. He's doing some big volume for that software.

Russ, NCNE
 
On the fiber channel and firewall note:

Fiber channel provides superior scale and performance over traditional SCSI. The bandwith is 100 megabytes per sec plus you actually build an &quot;infrastructure&quot; for storage rather than &quot;server islands&quot;. If you want to grow your storage system and maintain performance FC is the way to go. However these advantages come at a hefty price tag and only you can justify the expense based on your experience with these sites. A separate array is most definately the way to go, but whether you FC or SCSI attach it is your call. With an FC switch you can attach numerous hosts to your array and maintain performance. Again, hefty price...large returns.

This sounds like a decent web hosting environment and I'll stick to the cisco PIX recommendation (many models). Great performance, decent cost. I also love checkpoint firewall-1. Run this on a 400 Mhz Sun ultra 5 and again you get great performance/cost ratio.

What you're describing here sound like a static environment that needs to grow fast. For that reason alone I'm sticking to the above recommendations. SCALE baby, scale.🙂

<edit> blstriker - sorry man, it was very late and I was a little cranky. i'm chilled now. please accept my apology.
 
Thanks again for the help guys. Just wanted to pass along a little jewel that I found while browsing. It's a review for a RAID card that blows away Adaptec stuff Atto Express PCI UL3D

Also, what all components would I need to build an external RAID Array? Is there a typical CPU/Motherboard/Memory in a RAID array? Any tips on building one myself.
 
I might get chewed for this, but...

I live and die by compaq. Their storageworks products are just too easy to use and maintain. check their web page. A storage array is nothing more than power plus SCSI/FC interfaces plus backplane. Get one way bigger than you need now.

see here
 
If he's doing UBB, then that's dynamic material - UBB runs on cgi/perl. Get the fastest cpu's you can afford. Dual maybe quad even.

spidey07 - it's all cool. 😉
 
Back
Top