I don't disagree with any of this. Assuming you have the scale, I completely believe in segregating Exchange functions as much as possible (this approach always seems to work well in the windows world, even though it is much less necessary in the unix world). I had assumed the discussion of splitting up large private stores onto multiple servers involved first moving off connectors and such.
I'm not really advocating small servers. I was only commenting that I didn't think the uptime calculations were correct. I would never (except in a small, cost-constrained environment) install only a single mailbox server though. On this list, my preferred environment would be refered to as a "pairwise single-node cluster". Peter -----Original Message----- From: Soysal, Serdar [mailto:[EMAIL PROTECTED]] Sent: Wednesday, February 27, 2002 3:38 PM To: Exchange Discussions Subject: RE: Exchange 2,000 scalability question. I tend to agree with Ed. Segregating functions will improve your reliability (or perceived reliability) a lot more than spreading your user load. When you run dedicated mailbox servers, your downtime should be far less than it would be if you run multi-purpose servers (unless you use substandard hardware). Besides, since we're talking about Exchange 2000 here, you can still spread your users over multiple databases & storage groups to minimize impact of a database corruption. You really wouldn't run into any hardware related issues if you run good quality hardware and are religious about monitoring the health of your hardware. Serdar Soysal -----Original Message----- From: Reiss, Peter [mailto:[EMAIL PROTECTED]] Sent: Wednesday, February 27, 2002 3:07 PM To: Exchange Discussions Subject: RE: Exchange 2,000 scalability question. This logic seems faulty to me in two ways: 1) "How many users you are willing to have without e-mail when it fails" *is* a valid question. Even if the total person-minutes of downtimes stays constant (which I don't agree with either--see next point), having half of your users unable to work at one time may be very different (for your business) than having none of them able to work. My office only has about 200 mailboxes, but we have very high uptime requirements for our business (we are a trading firm), so I would never consider putting all of these mailboxes on a single server (they are on 2 currently). At least if one server goes down, half the users are unaffected. 2) I think your person-minutes calculation is wrong. More precisely, I think it is incorrect to assume that if a 2000-user server has 99.95% uptime, that two 1000-user servers each have 99.95% uptime. Downtime is calculated as frequency of "downtime" events (unplanned or planned) and the duration of those events. Assuming a scenario that requires a restoral, consistency check, or some other time-consuming Exchange process, the number of users (and hence size of the store) are a major factor in determining the length of the downtime incident. So in effect, by reducing the number of users on a server, you may be reducing the total downtime (time-to-recovery). So if time-to-recovery of a 1000-person server is 60% of time-to-recovery of a 2000-person server, then the total person-minutes of downtime in the 2 1000-person servers is 2 * 1000 * .0005 * .6 = 315,360 (not 525,600). [If a lot of the downtime is planned and duration is irrespective of server-size, then this effect is smaller.] The point about reducing inter-server communications is probably valid, and it certainly is true that having very many small servers is probably a bad solution from a reliability standpoint (management is less likely to be as good). Peter -----Original Message----- From: Ed Crowley [mailto:[EMAIL PROTECTED]] Sent: Wednesday, February 20, 2002 12:11 PM To: Exchange Discussions Subject: RE: Exchange 2,000 scalability question. The argument that "It's...how many users you are willing to have without e-mail when it fails..." is based upon faulty logic. My rebuttal argument is thus: Say you have 2,000 users on one server. It stays up, say, 99.95% of the time. Then you can expect 525,600 person-minutes of downtime per year (262.8 minutes times 2,000 users). Say you have 1,000 users on each of two servers. Each stays up, say, 99.95% of the time. Then on each box you can expect 262,800 person-minutes of downtime per year for each machine, the aggregate being 525,600 person-minutes. Each scenario has the same amount of person-minutes of downtime. The argument that more mailbox servers increases reliability would logically extend to the point where maximum reliability is achieved by giving each user his own Exchange mailbox server. Of course, it is obvious 2,000 servers for 2,000 users will not increase reliability. A failure of a single server will only affect one user, but you now have 2,000 such servers failing occasionally. One could argue that having fewer mailbox servers will actually improve reliability because there is less inter-server communication. I do buy the argument that splitting off functions to separate servers can improve reliability because these functions can cause the mailbox server to fail. Overall system reliability may not change, but perceived reliability will be higher. For example, if you have problems with an SMTP Connector, you might have to take an outage on the server to fix it. If the SMTP Connector is on a separate box, it's likely that a brief failure won't even be noticed by users, but if you have to cycle the mailbox server, far more will notice. Ed Crowley MCSE+Internet MVP kcCC+I Tech Consultant Compaq Computer Corporation Protecting the world from PSTs and Bricked Backups! -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] On Behalf Of Stewart Jump Sent: Wednesday, February 20, 2002 3:58 AM To: Exchange Discussions Subject: RE: Exchange 2,000 scalability question. Its not so much how many users a box can support but how many users you are willing to have without email when it fails and how long you are wiling to sit there while restoring the server. 3000 users @2MB is only 60GB. Assuming you have an LTO tape drive (100GB native + compression at ?:1) then you could probably get 6000 users on a server and back it up to one tape, which makes running a full backup every night a lot easier. Our test suggest a restore time of around 800 to 1000 MB / Min so restore times aren't to long either. As for how much CPU and RAM you need it is really a guess as you can sit in the lab for ages doing LoadSim tests, and still get it wrong in the real world due to changing patterns of usage with the new features. Our test's with the latest LoadSim version on a system with dual CPU 700Mhz PIII with 1GB of ram suggested it was the disks for the stores that where limiting the system when trying to simulate 3000 users, and not the log drive even when running 4 storage groups Log files onto 1 dedicated mirrored pair (1). Regards Stewart Jump (1) I know its not recommended but it does mean the log pair is on a different SCSI bus to the Store drives and the performance doesn't seem to suffer until you run a backup during a LoadSim run (2) (2) This why you need to do your own tests as the "Rules" can be bent and still get a decent configuration. -----Original Message----- From: Pfefferkorn, Pete (PFEFFEPE) [mailto:[EMAIL PROTECTED]] Sent: 19 February 2002 19:13 To: Exchange Discussions Subject: Exchange 2,000 scalability question. Well, First let me apologize for the vagueness of this question. Just starting to read up on Exchange 2000. I've been asked by management to spec out the possibility of deploying Exchange 2,000 to all users on our campus (40,000 users). I'm reading Tony Redmond's book on Exchange 2000 and it mentions on page 21 that on 5.5 maximum user community was in reality about 3,000 per server for a 8 way. I know that really depends on the server setup, user load, etc. Our Dell 8650's with 2-550mhz processors and 2 gig of memory is handling about 3,000 users per server with 20 meg quotas per mailbox. They are active accounts (light to medium), so the 3,000 seems to match up since our system CPU utilization averaging about 80-90% during peak times. On the same page it states that Exchange 2000 states about 10,000 for a single node cluster. I'm wondering if anyone had opinions on this issue. What is different about Exchange 2000 that allows for more users per system. Is this do to coding changes in the store.exe or is it based on multiple stores per server? Given we can get about 3,000 users per Exchange 5.5, I'm wondering what the actual limit would be for Exchange 2000 per server. Pete Pfefferkorn Senior Systems Engineer/Mail Administrator University of Cincinnati 51 Goodman Street Cincinnati, OH 45221 Phone - (513) 556-9076 Fax - (513) 556-2042 _________________________________________________________________ List posting FAQ: http://www.swinc.com/resource/exch_faq.htm Archives: http://www.swynk.com/sitesearch/search.asp To unsubscribe: mailto:[EMAIL PROTECTED] Exchange List admin: [EMAIL PROTECTED] _________________________________________________________________ List posting FAQ: http://www.swinc.com/resource/exch_faq.htm Archives: http://www.swynk.com/sitesearch/search.asp To unsubscribe: mailto:[EMAIL PROTECTED] Exchange List admin: [EMAIL PROTECTED] _________________________________________________________________ List posting FAQ: http://www.swinc.com/resource/exch_faq.htm Archives: http://www.swynk.com/sitesearch/search.asp To unsubscribe: mailto:[EMAIL PROTECTED] Exchange List admin: [EMAIL PROTECTED]
