hi, i wanted to check with some of you smart folks before moving ahead. can you please tell me if this look alright? btw, this is for a renderfarm. thanks. ..10/100nic--switch--pc's & firewall--net ./ Fileserver .\ ..gig-e nic--gigabit switch---PC"A" ............./..../...\....\ ............/..../.....\....\ 10/100switch..switch switch..switch ..........|.....|.......|......|... ......8nodes..8nodes..8nodes..8nodes sorry for the "."'s. it didn't lay out properly. ok, PC"A" is a machine that needs to eat up alot of throughput, reading and writing large video files to the fileserver. the 32 nodes are segmented on 10/100 8 port cheap switches connected to a 5 port gigabit switch. the 1000bt nic in the fileserver is sitting in a 32bit slot. I was wondering if this layout made sense and if I should use my fileserver's 3 free pci slots with 100bt nics instead. thanks. Alpay

Gigabit isn't going to do you much since you are only having only one pc doing major uploads. You can get a HP4000 switch with 48 ports which will have better bandwidth then what you propose with multiple switches cascaded from the gigabit switch. You can put in a gigabit interface in this switch and connect to a gigabit card in the server if you want to optimize server thru-put. If you want to get fancy you can put in two 100mb cards [intel] in the pcA workstation and do adapter teaming. This will give you 400mb bandwidth [2x200mbps full duplex=400mbps] Couple of concepts to share with you. Gigabit does not mean faster. It means more bandwidth. This means you can put more on the network hwy without saturating the bandwidth. Network saturation is what "slows" the network down. Second concept is more server nics doesn't make the server faster. Faster disk subsystem [like raid10] and more ram in the server will improve performance. Even if you did adapter teaming with two gigabit nics [4000mbps] the thru-put is still limited to the disk subsystem. In other words you can only read and write at a set rate which is uneffected by the network bandwidth.

Wanderer, your info was more useful than most I've gotten on a couple of forums. I should say this though... All the nodes are moving a lot of data TO the fileserver, while PC"A" will really be doing constant I/O to the FS. I thought that segmenting things this way would relieve bottlenecks trying to get to the FS. The FS is in the middle, btw, because i thought of multihoming the file server so that the rendernodes could be on their own network and also maybe be firewalled from the regular pc's which might catch some nasties from the web. I'd appreciate if you could tell me how you might set things up. I have not purchased any of the switches yet, I was looking for this kind of advice first. thanks again.

If you are expecting to have high IO you don't want to run the internet thru the server. If you get a Sonicwall firewall you can have it do the nasties scan [number of features by subscription]. I would keep as many services off the server I could to maximize its service. Have the firewall provide dhcp [with your server having a static address- you can do this with dhcp -reserved address]. Your disk subsystem is very important. Adaptec raid card 2200s is a good choice as well as scsi drives [15K rpms. Mirror the OS and raid10 [never raid5!!!] the data. Consider a Hot spare for failover. In the case of segmenation you will actually add latency with multiple switches since the pc request has to first be shared with pcs on the same switch, then it goes to the backbone switch which is shared with the server and other switches. You may want to consider adapter teaming in the pcs or even gigabit [now your talking some money]which would translate to a higher end switch. You really want the fewest devices between the pc and the server to optimize your thru put.

The internet does not go through the server, the "nodes" are for a renderfarm, they don't need the net, just the fileserver. the firewall I have listed in my diagram is my dhcp server as well. using smoothwall on an old athlon machine for that. works fantastic. Why should I stay away from raid5? I have a Promise Sata raid controller in there now set up for raid5 with a terabyte of 7200rpm western digitals. I forget the results of my read/write tests but it was quite fast. It's sitting in a 3.2 p4 with a gig of ram and there is 128mb cache on the raid controller. I was considering adding another terbayte with the same Promise card and striping those. Good point with the added latency on the segments. I will be buying a a 24port 10/100 procurve with 2 gigabit uplinks which will feed the multihomed fileserver. I think thats my solution right there. Thanks for the help guys. Please do let me know if this last paragragh is the wrong way to go however. I'm excited to get set up now!