IT Employment

Ask something
Discussion Locked

How many servers do I need?

0 helpful
4 replies

by hig

Closed to new repliesThis discussion is closed. It stays readable, but it takes no new replies.

There is a discussion going on among some IT folks I know concerning how many servers are best for a certain scenario and I wanted to get the collective wisdom of this group. This is the scenario: XYZ Engineering company has a large regional HQ office that is home to approximately 125 engineers. The engineers are using softwares that generate large data files, this software includes CADD software, Modeling software, and Structural Analysis software as well as a standard support staff using the MS Office suite. 4 regional branch offices also have connectivity to these servers for data sharing among the offices. The office currently has 3 data servers with each server serving a particular logical area, for example one server houses CADD files, one houses modeling data, one houses office data, etc.
One side of this debate believes that it would be best to consolidate all the data on one server. This reduces hardware costs and provides one backup point and less confusion among data points. The other side of this debate believes that by distributing the data among several servers that there is a natural load balancing of data IO at the disk array and the network interface and that several points of failure exist rather than one. In other words, if all the data were on one server and this server failed everyone in the office is out of work until it is fixed, if there are three servers and one fails, only one third of the staff is out of work. All advice, comments, and snide remarks are welcome:
Thanks,
David

4 replies

by chris910

the mentality of your organization.
In my situation (40 servers,18 sites, 5500 workstations) the Tech's thought the best solution was to consolidate a portion our servers as we upgraded our networks to gigabit backbones and T1 links to our remote sites. What we had in mind was to reduce our server count from 40 to 28; make use of the faster network and larger drives to handle the more routine applications; while retaining our specialized workgroup servers in key locations to reduce wan traffic; allowing for some autonomy; and for projects with special funding.
Here is the rub:
Management was having fitts about the initial cost of buying the servers. The cost of increased Processor, ram, drivespace, network cards etc was driving the cost to over 2 times what we had been paying for our standard workgroup servers. Management never got past the per-unit-cost and as a result we replaced servers on a one-for-one basis. Today we maintain more servers with a lesser level of support than we are comfortable with.

As to your specific situation I would give some thought to the following things before deciding to consolidate your CADD and modeling Servers. (I am making a basic assumption that the CADD and Modeling files are fairly large)
-- at quitting time (lunch time, starting time) will there be enough network bandwidth for your employees in both the CADD and modeling groups to save their work simultaineously.
-- Do you need to upgrade yor network bacbone to accomodate the greater traffic concentrated at the new server.

by hig

Thanks Chris910, YIKES! 40 servers and 5500 workstations, you folks are busy I'll bet. I empathize with your experience with the bean counters. Your assumption is correct that the CADD and modeling files are indeed very large. GIS databases can quickly exceed 50GB's per each. Lots of nodes upon which calcs are performed and well you get the picture. The same goes for the CADD files but to a lesser degree. It ahs been suggested that putting in a beefy server with two gigabit over fiber cards would certainly increase the bandwidth at the server but the situation still has 125 engineers trying to open or close their data files over these two pipes at one time at least twice a day. In other words 125 workstations at 100megabits per station does not equal 2 server nic's at 1000 megabits per card by a long shot. The network backbone in this situation is a Cisco 6509 switch which is in place and should provide all the network backbone one could reasonably throw at it. The two sticking issues in this discussion remain network bandwidth at a single server and the issue of one point of failure. It has been suggested that if the server failed, however unlikely given the reliablity of currently available hardware, that 125 engineers will be out of work for the 4 hour vendor response time. This cost would far exceed the cost additional server hardware. What do you think?

by this4eagle

for what it's worth I was N/A with a civil engineering company for several years and had about 50-60 production servers and 3000 nodes .... or something close to that. Anyway, we went through two major infrastrcuture upgrades and redesign from lan to wan from fast, switched,gig, fiber, etc. Your cadd files are and will be much too large for acceptable margins in daily management w/ collapsing or single server route. There will be noticeable degradation, choking bandwidth & server proc. etc.(which promotes downtime), and downtime would be unexceptable expense per cadd w/s to your CFO especially coming from the IT dept. which is not a revenue center and is already an expense. I suggest to distribute server farm by category and/or % of usage, with as much processing speed and fat pipes your company can afford. We did so and kept unscheduled downtime averaging less than 2% monthly. Simple network strategy once again is best. Don't re-invent the wheel your environment is easier than most. I hope I helped and was somewhat clear ... had a long day.
Good luck, & God Bless!

by pipe guy

Many firms go through the same sorts of debates. In engineering type firms, you can't afford any downtime, because everyone notices right away. The four hour response time for 125 people could cost you your job in some firms. And depending on the situation and the type of fix that is required... 4 hours isn't a long time to be down. That being said, if you are just going to calculate data and bandwidth requirements, you might get away with the server consolidation model. But even if you installed a Netware server and a huge whack of scsi hard drives to handle the sort of print/file storage requirments that you will have, you will still have bottlenecks during some times of the day when everyone is opening up their 1gb+ files. And engineers will complain if it gets even a little bit slow.
If you segregate your services based on departmental needs and a have a seperate server for each (1/3 for each) you will never have everyone down at once... maybe... unless the problem is a security flaw that forces you to shut everything down to fix the problem. But it all depends on the hardware you have chosen and the OS that you are planning to run. Unless you absolutely have to run a Windows server for some services like "voicemail" or if you run application software that must run on a MS box, you might consider Netware 6.5 for small business. It lets you have up to 100 similtaneous users logged on per server and since the kernel is a lot more efficient than Server 2003, you will have much better efficiency for file transfer and storage. Faster by about 60% on the same hardware. Another huge advantage of the Netware product (ever since version 6) is that for many of the file transfers you only have to copy the changes that you have made to the file. In other words... the engineer has downloaded a 500 mb file for a particular project... he makes a minor change... and now he doesn't actually have to upload the whole 500 meg file or larger back to the server... only the actual changes that were made have to be transfered over the wire. So when he "saves as"... and he keeps the same file name... only the changes to the file get transfered. This ensures that the efficiency is greatly increased over having to copy the entire file. Multiply this 100 times and you could easilly do it with one netware small business server. But then again I would probably recommend that you have at least two servers and in the event of a hardware failure... you wouldn't lose the downtime of the entire organization. With the quality of a lot of the hardware being sold these days and the Dell ads for 700.00 servers, I would buy the "departmental" quality servers with U320 or faster Scsi hard drives. But you realize that even the Gigabit ethernet backbone isn't going to keep up with a good P3 1.2 ghz motherboard let alone a 533 front side bus P4 processor. So your infrastructure is going to require some serious investment to keep up. The server may be the cheapest part of the network. You'll want to have some 64 bit slots and dual or quad port nics in the server as well as some clean connections to the other servers. For example if you have a dedicated switch just for traffic between the servers and have separate NIC ports just for this... you will improve the speed for your Disaster recovery software and backup processes, assuming you have only one server doing the tape backups. (possibly in a fireproof room) But can be a very complicated process designing the network to have maximum efficiency and you will always have a slow link somewhere that causes you grief for many months. Until some diagnostic software finds the defective connection. At least that has been my experience. It is an interesting proposition and will take at least a few weeks to design the "perfect" model.

This discussion is closed to new replies.