John Haxby wrote:
This is far too open-ended a question to answer properly. But I can give some general advice based on the sizing stuff we use for Contact. I think sizing information for Contact may be available on the web site, I don't remember.
Anyway, for web, database, e-mail and file servers the overwhelming need is for fast I/O. If you're dealing with less than about 100 active users then just go to your favourite hardware emporium and buy the cheapest machine they have there. The disk speed will be fine. Memory will be fine (256M, I expect), CPU speed will be hopelessly over the top. Of course, you won't get a mirrored disk for that so you'll suffer when the disk breaks.
After that we reckon on something like 10 spindles, 1Gb and 1 processor per thousand users (something around there anyway). An active user is someone who is actually logged on and busy -- our profile is based on 20% of configured users being active. This is fine in a coporate environment. For consumer users (who use POP for e-mail rather than IMAP or MAPI), it's not even as much as 1%. You need to know your workload. And you need to know how to measure system performance and emulate that workload.
What the heck?
We have 7000 pop3 accounts and over 1000 domains on a Compaq Proliant with a 650MHz PIII and 256 Mb RAM, we use a compaq smart2 raid controller and 4 19GB drives as two 19GB mirrored logical drives {one for the mail spool}. The system runs at an average load of 0.40 . The server currently acts as Primary DNS, POP3, SMTP {Outbound and from cleaned MX}, and WebMail {IMAP based with folders disabled}. We handle MX delivered mail through another system to offload virus and spam filtering, then forward the "clean" mail to that server. We also have secondary DNS and SMTP {MX} servers, and multiple Web, DB and other servers.
I would expect that a good rule of thumb is to not go too crazy on expensive hardware, but get good hardware for primary and critical systems. Depending on how critical the system is will be how you should determine your costs. If you can afford a 1/2 hour of downtime to replace a failed IDE drive from a mirrored system, then you don't need to spend $6000.00+ on a brand name server class machine with redundant power supplies and expensive RAID controllers with expensive SCSI drives and with hot standby failover capabilities. But if you have a mission critical 24x7 system that can never fail or loose any data, then you probably need more advanced hardware, software and support than you will get in this list.
When I worked at Xerox, there were two questions that had to answer for any changes to projects or procedures:
1) What is the cost of conformance?
2) What is the cost of non-conformance?
In other words, if the cost for your system is tens of thousands per year and the cost of having another system that may fail is the loss of your business, then buy the system. But if you can afford a little down time and a possibly few missed emails then the cost for the system is too high.
The first thing you need to do is answer both of those questions, then you can make a qualified decision, anything else is speculation.
Having said all that, for a few hundred users, get a Dell Poweredge 1650 two CPUs with about 1G memory, and a megaraid controller. You'll need a SCSI disk array and about half a dozen disks (mirrored and striped, RAID5 doesn't have the write performance you'll need for mail and file serving. It's a noisy machine, but it comes with maintenance and someone who knows how to look after it. Make sure you get an array with similar maintenance. Get a spare disk (one is enough). Get a nice SCSI tape drive, a DLT or something to do backups. Don't go for cheap no-name hardware, you won't be able to get it fixed when it goes wrong. Run RedHat ES or AS (depending on whether you want hardware failover).
jch
M. Fioretti wrote:
Hello,
not specific to shrike, or even Red Hat for that matter, but probably of interest to many members of this list, and many others surely have the expertise to answer:
In the x86 realm, what is the perfect HW architecture for a box which is only ONE of
a web server a database server an email server a file server
In other words, assuming money is no object, where should it go first to give top performances in each of the cases above, and how and why they differ from each other?
I'm looking for answers like "for this kind of server you should buy first of all a CPU with at least this much of L2 cache because..." or "for this other server the most important thing is this kind of RAM, or this (set of) hard disk, because..."
Any feedback or pointers to online resources is appreciated. Thank you in advance,
Marco Fioretti
-- Guy Fraser Network Administrator The Internet Centre 780-450-6787 , 1-888-450-6787
There is a fine line between genius and lunacy, fear not, walk the line with pride. Not all things will end up as you wanted, but you will certainly discover things the meek and timid will miss out on.
-- Shrike-list mailing list Shrike-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/shrike-list
