RE: [sniffer] Moving Sniffer to Declude/SmarterMail
Pete OK, I now have much more information on this problem with Declude/Sniffer/SmarterMail. It seems the current version of Declude does not have an Overflow Directory for SmarterMail, which therefore allows unlimited Declude processes to be spawned at any time. At our peak we were seeing a surge of more than 1,000 declude.exe instances running at the same time! This of course flattened the server, and seems the reason why Sniffer was dropping out of its perpetual mode, unfortunately compounding the problem when the server had least resources. On speaking with Declude, thankfully they have been working on a version to control the number of processes running, and have let us have a BETA version which allows us to set the number of processes to a setting of our choice - we have it at 30 and it's working fine! One thing we did whilst in the middle of this was to move all the log and spool files to a standalone disk instead of the RAID5 array for the main server, and we have seen a real reduction in the physical disk queue lengths, which, under significant load, helps. Worth knowing. So the migration is complete with all users running as normal on SmarterMail instead of iMail. Hope some people can learn from our pain! Nick -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Pete McNeil Sent: 14 March 2005 19:23 To: Nick Marshall Subject: Re: [sniffer] Moving Sniffer to Declude/SmarterMail On Monday, March 14, 2005, 12:47:33 PM, Nick wrote: NM Hi there NM We've just undergone a migration of a 1,000 domain iMail server to NM SmarterMail (for obvious reasons!), and using Declude and Sniffer on NM the new system. NM However, occasionally we see Sniffer jumping out of its perpetual NM mode by spawning thousands of threads which obviously make the NM server slow down considerably. Also, at this time, the Sniffer NM folder fills up with the equivalent temporary files. NM On changing the location of Sniffer in the Declude Global file and NM allowing the threads to disappear, when Sniffer is re-engaged it NM correctly goes back to perpetual mode working from the Service instance of Sniffer. NM Can any of the above be explained? This is extremely unusual and I've not heard of this kind of thing before. The closest connection I have seen is that on one very heavily loaded system (on Linux) a change in the rulebase file caused the persistent instance to pause for a few minutes while the message instances of SNF became impatient --- however in this case the system always recovered on it's own. In your case do you restart the persistent instance or does the system simply recover once you re-enable the clients? Please post the contents of your .stat file and let me know if these are typical values. Does your system have one or two CPUs? Hyperthreading? Thanks, _M This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html -- [This e-mail was scanned for viruses by Giacom Anti-Virus] -- [This e-mail was scanned for viruses by Giacom Anti-Virus] This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
RE: [sniffer] Moving Sniffer to Declude/SmarterMail
Thanks John - I didn't know that, but it would explain things... Nick -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of John Tolmachoff (Lists) Sent: 16 March 2005 14:40 To: sniffer@SortMonster.com Subject: RE: [sniffer] Moving Sniffer to Declude/SmarterMail One thing we did whilst in the middle of this was to move all the log and spool files to a standalone disk instead of the RAID5 array for the main server, and we have seen a real reduction in the physical disk queue lengths, which, under significant load, helps. Worth knowing. Nick It is a well known and published fact (on the Imail list) that RAID5 should never ever be used for the spool directory or any other directory that has a high write activity. This is basic physics. RAID5 should really only be used for high read activity only, such as databases where most of the writing is done to transaction (log) files and at spaced intervals those transactions are committed to the database. RAID1 or even RAID0+1 is best for the spool and logs. John Tolmachoff Engineer/Consultant/Owner eServices For You This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html -- [This e-mail was scanned for viruses by Giacom Anti-Virus] -- [This e-mail was scanned for viruses by Giacom Anti-Virus] This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
Re[2]: [sniffer] Moving Sniffer to Declude/SmarterMail
On Wednesday, March 16, 2005, 9:01:34 AM, Nick wrote: NM Pete NM OK, I now have much more information on this problem with NM Declude/Sniffer/SmarterMail. NM It seems the current version of Declude does not have an Overflow Directory NM for SmarterMail, which therefore allows unlimited Declude processes to be NM spawned at any time. At our peak we were seeing a surge of more than 1,000 NM declude.exe instances running at the same time! This of course flattened the NM server, and seems the reason why Sniffer was dropping out of its perpetual NM mode, unfortunately compounding the problem when the server had least NM resources. Thanks for this --- I think this proves one of my theories on the problem. This is what I think happened to SNF... A burst of 1000 or more active requests arrives which is more than the server can really handle at one time. The persistent server queues (internally) all of the jobs and begins processing them all at once. As a result, it does not report to the .stat file for an extended period. Also, due to the very large number of jobs, many of the client instances do not hear back from the server instance until their maximum wait time has expired. As a result, they abandon the wait and begin processing the messages themselves, each one now loading the rulebase. The clients loading the rulebase individually further slows the server and accelerates the problem until an insurmountable backlog is in place. Once the backlog is cleared (requiring manual intervention) the persistent instance (which has not died) is able to respond to client requests quickly enough so that they do not give up and so the system operates normally. --- If this fits the profile (and I think it does) then it would be possible to adjust SNF with an alternate fail-safe mode which would solve the problem at the expense of letting more spam through. Specifically, once a particular threshold is reached then the clients would abandon a job and fail safe rather than loading the rulebase and processing the message. This would cause spam to leak but it would also provide for a more rapid, automatic recovery. Think of it as a safety pressure valve on a boiler. It's not a great solution, but it's better than an explosion. The best answer is proper overflow control and since Declude is working on that I'm inclined to let that solution develop. Not only because it's a better solution, but also because the safety valve mechanism on SNF doesn't solve the whole problem - for example virus scanning would still potentially cause the same problem along with other things that might be running in Declude. Thanks Nick! _M This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
RE: [sniffer] Moving Sniffer to Declude/SmarterMail
John, It is a well known and published fact (on the Imail list) that RAID5 should never ever be used for the spool directory or any other directory that has a high write activity. This is basic physics. RAID5 should really only be used for high read activity only, such as databases where most of the writing is done to transaction (log) files and at spaced intervals those transactions are committed to the database. RAID1 or even RAID0+1 is best for the spool and logs. I guess this is going against what I think should be happening. In a RAID 5 array the write to the drives is broken into many smaller writes along with the data protection/CRC info and then those writes are written to different drives. It seems to me that it should be faster to do a bunch of small writes rather than 1 big write. What am I missing? Goran Jovanovic The LAN Shoppe This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
RE: [sniffer] RAID level for spool
Even if you break it into smaller blocks, you still need to transfer the data to the controller, then the controller has to employ overhead to break up the block, create the parity information, determine the location for each block, etc. With RAID-1 the controller can just write through and duplicate the operation as is on the second bank. http://www.acnc.com/04_01_05.html vs. http://www.acnc.com/04_01_01.html RAID-1 has less overhead during writing. Since the spool folder probably has a 1:1 read/write ratio - it is sensitive to write performance. RAID-5 works well for write once - read many times applications, such as file and database servers. Best Regards Andy Schmidt Phone: +1 201 934-3414 x20 (Business) Fax:+1 201 934-9206 -Original Message- From: Goran Jovanovic Sent: Wednesday, March 16, 2005 11:26 AM I guess this is going against what I think should be happening. In a RAID 5 array the write to the drives is broken into many smaller writes along with the data protection/CRC info and then those writes are written to different drives. It seems to me that it should be faster to do a bunch of small writes rather than 1 big write. This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
RE: [sniffer] Moving Sniffer to Declude/SmarterMail
-Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Pete McNeil Sent: 16. marts 2005 17:43 Writing data to a raid 5 takes x+y+z amount of work where y is described above and z is calculating a CRC stripe which must now also be saved to a hard drive. And from my understanding, calculating the CRC requires knowing both the data in the sector overwritten and the previous CRC data, and thus 2 reads. So thats 2 reads, and 2 writes to write a single block of data. Regards, Kaj This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
RE: Re[2]: [sniffer] Moving Sniffer to Declude/SmarterMail
OK that is for hardware level RAID. I had thought that you would offset the extra processing time by being able to write less to each drive. Now does anyone know how much overhead Windows 2000/2003 software RAID 1 on dynamic disks produces over hardware level RAID 1? I am assuming it would be substantial. Goran Jovanovic The LAN Shoppe -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Pete McNeil Sent: Wednesday, March 16, 2005 11:43 AM To: Goran Jovanovic Subject: Re[2]: [sniffer] Moving Sniffer to Declude/SmarterMail On Wednesday, March 16, 2005, 11:25:46 AM, Goran wrote: snip/ GJ I guess this is going against what I think should be happening. In a GJ RAID 5 array the write to the drives is broken into many smaller writes GJ along with the data protection/CRC info and then those writes are GJ written to different drives. It seems to me that it should be faster to GJ do a bunch of small writes rather than 1 big write. GJ What am I missing? Writing data to a single hard drive takes x amount of work. Writing data to more than one drive takes x+y amount of work where y is breaking up the data into chunks. Writing data to a raid 5 takes x+y+z amount of work where y is described above and z is calculating a CRC stripe which must now also be saved to a hard drive. So, writing to raid5 is relatively very expensive compared to writing to a plain old hard drive, or a less complex raid (such as mirroring). IMO, the best strategy for email servers is to use an ordinary, single fast HD for all spool operations, and place mailboxes on a raid 1 or raid 10. Hope this helps, _M This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
RE: [sniffer] RAID Levels for Spool Folder
Uh, sorry, I had thought that discussion was RAID-5 vs. RAID-1? If someone is running RAID-5, I assume that it's hardware based. If so, then that person could use the same hardware to configure a RAID-1 array instead - so why even bother with software RAID then? If the discussions is software RAID-1 vs. no-raid, then the answer is: Sure, software RAID is a cost effective solution if the system has sufficient head-room to deal with whatever possible overhead that may cause. However, if we are talking about a machine that is already taxed, then I would suggest plugging in a RAID controller instead of adding software RAID to the mix. I have several (older) systems running Windows 2000 RAID-1. At least ONE of the servers I later upgraded to Hardware RAID. I can't say that I've noticed any difference (but then again, I have not run benchmarks and the server was not really taxed before either.) From what I understand, there are many factors involved and it much depends on your systems configuration. CPU availability is critical. A server that is already CPU taxed may suffer if software RAID is added. Having the drives split on two SCSI controllers should also help with software RAID-1. Doing software RAID-1 with a master/slave ATA drive, however, may slow things down. There may not be a single answer... Best Regards Andy -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Goran Jovanovic Sent: Wednesday, March 16, 2005 02:05 PM To: sniffer@SortMonster.com Subject: RE: Re[2]: [sniffer] Moving Sniffer to Declude/SmarterMail OK that is for hardware level RAID. I had thought that you would offset the extra processing time by being able to write less to each drive. Now does anyone know how much overhead Windows 2000/2003 software RAID 1 on dynamic disks produces over hardware level RAID 1? This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
RE: Re[2]: [sniffer] Moving Sniffer to Declude/SmarterMail
Now does anyone know how much overhead Windows 2000/2003 software RAID 1 on dynamic disks produces over hardware level RAID 1? I am assuming it would be substantial. I have never noticed an issue, and I would only assume there would be an issue in higher end databases or where the CPU was already being tasked and near or at saturation by other processes. John Tolmachoff Engineer/Consultant/Owner eServices For You - This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
Re: [sniffer] RAID Levels for Spool Folder
IMO, Software RAID is not the way to go on a busy machine. You will save a measurable amount of overhead by going with hardware based RAID of any sort since the controller should handle the processes associated with the RAID. Note that this isn't the case with inexpensive RAID controllers such as the cheaper IDE and SATA controllers which still place a fair burden on the OS/processor. True RAID cards also offer additional cache which can speed up the performance on reads, and also on writes if you are battery backed up (otherwise don't use write caching because you could lose or corrupt data during a power outage). There's also several common misconceptions about what is proper to do for a mail server. RAID 5 is the best choice under almost all conditions. The trick here is that while RAID 10 offers both redundancy in mirroring and speed in striping, most servers have a limited amount of space for disks. So a server with 6 disks will operate with the speed of 3 disks spanned in a RAID 10 configuration, but 6 disks in RAID 5 will operate as 5 disks spanned plus a little bit of overhead, though not nearly enough so that it falls short of the performance of just 3 disks in a simple span. Therefore RAID 5 should be the default choice for speed in such an environment. Another misconception is that data is always striped in RAID 0 or RAID 5. This depends on the file size and the stripe size. Most stripes are 64 KB (configurable in most setups). If you have some form of striping for your spool drive, most messages fall far under 64 KB and will only get written to one disk (CRC will also get written in RAID 5). Therefore for a spool folder, RAID 5 with 3 drives (the minimum), will perform rather closely to RAID 5 with 10 drives since most files will only land on one disk (with the other corresponding stripes containing no data). The MFT however for a drive with a lot of files will grow to be quite large and benefits from having multiple disks, and opening very large files such as logs will also benefit from having many disks. There is also an advantage to seek times when having multiple disks, especially if you keep your partitions sized small for performance. I've run a dual processor 3.06 Ghz server with both 6 Seagate 15,000 RPM drives in RAID 5 and the same with 3 Seagate 10,000 RPM drives in RAID 5 running on a less capable controller, and there was no impact on performance while the server was handling over 125,000 unique messages a day. The only noticeable difference was the time it would take to open a 500 MB log file, or the time it would take to enumerate the file names from the MFT on a partition that contained tens of thousands of files in the root. It seems quite apparent that with modern processors, even in dual processor configurations, that you will run out of CUP cycles long before you run out of disk I/O in a well managed RAID 5, 3 drive configuration on an IMail/Declude/Sniffer server. Take note that the log files for Declude, Sniffer and IMail all become massively fragmented, and if you don't have a process to remove these from active partitions on your server or defragment them, then performance will be severely impacted. I run a job hourly that copies all such logs to a different partition and combines them with older chucks from that day and then zips them nightly. The process of moving them to another logical drive removes the fragmentation and that helps to ensure that the spool or mailbox partitions/folders don't also become heavily fragmented, which is a big performance hit. Opening up a heavily fragmented 500 MB Declude log file is excruciatingly slow. If I kept the logs in my spool folder without taking action to handle the fragmentation, each Declude log file would reach over 100,000 fragments a day, and that's a lot of seek time. I would recommend just going to RAID 5 for everything, and buying an LSI or Adaptec card if doing SCSI, or one of the same including 3Ware if doing SATA or IDE. Based on my personal experience, I don't believe that you need to go over the top with anything, just the cheapest brand name card that can handle RAID 5 will do, even in a zero-channel configuration if your motherboard supports it. Ultra 160 SCSI RAID cards will also work just fine for all but the most demanding applications these days, so don't be afraid to pick one of the older models up from eBay. Also pay attention to the drives themselves, SCSI drives are made to be better and more dependable than most IDE or SATA drives, and the faster RPM's mean faster seek times as well, and different brands also require less CPU overhead. Only Western Digital seems willing to produce a server-class SATA drive, and this is because they are the only SATA drive maker that doesn't have a SCSI line that might be impacted. SCSI as a protocol also offers some things that still aren't commonly implemented in SATA that will improve performance in a RAID configuration, though that will change over time.
Re[4]: [sniffer] Moving Sniffer to Declude/SmarterMail
On Wednesday, March 16, 2005, 2:05:00 PM, Goran wrote: GJ OK that is for hardware level RAID. I had thought that you would offset GJ the extra processing time by being able to write less to each drive. GJ Now does anyone know how much overhead Windows 2000/2003 software RAID 1 GJ on dynamic disks produces over hardware level RAID 1? GJ I am assuming it would be substantial. This is difficult to quantify, but in my limited experience the difference in load between software RAID1 and hardware RAIDx is approximately the same as the difference between using IDE drives and using SCSI drives. SW RAID1 is not hard on the CPU except that it has to coordinate the write operations etc which leads to some extra queuing work/delays. With a good hardware based RAIDx the effect is that this coordination is handed off to the controller so the CPU spends more time doing actual work and less on managing the drive subsystem. This is roughly similar to the performance gain you get by going to a decent SCSI based system since SCSI systems have higher level communications protocols that generally allow the CPU to get on with other business in much the same way. In the end these choices are all about how much money you have and your performance goals. _M This E-Mail came from the Message Sniffer mailing list. For information and (un)subscription instructions go to http://www.sortmonster.com/MessageSniffer/Help/Help.html
RE: [sniffer] RAID Levels for Spool Folder
Matt, I think that you sort of answered the question that I did not really ask. I was really trying to get information on the different performance levels for of S/W vs H/W RAID for an ideal scanning only box. So let me try this out and people can comment All SCSI 15K drives with HW RAID controller 2 x 36 GB drives R1 on first channel (36 GB usable) C Windows 10 GB D IMAIL/Smartermail/Declude files/Declude filters per domain configs/banned files (5 days only) 20 GB P Page volume 3 GB 3 x 36 GB drives R5 on second channel (72 GB usable) L Logs for JM, Virus, IMAIL/SmarterMail, Sniffer, invURIBL, et al 10 GB S Storage for all daily logs 60 GB 1 x 36 GB Hot Spare drive From what we have discussed here drive L will get hit a lot. If you create a process that Matt is describing to move the active logs from L to S you should not worry about running out of space on the L drive. Now looking back I am not sure if I have crafted this well since the SPOOL files for IMAIL will end up on D. Is there a way to move them for Smartermail as there does not seem to be a way to move them in IMail? The good part of this config is that the spool files which have a lot of read/write are on a different volume/channel from the other log files. I am not sure what amount of space you should allocate to a server that would process 100,000+ messages a day? Anyone have comments on this config. Thanx Goran Jovanovic The LAN Shoppe From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Matt Sent: Wednesday, March 16, 2005 3:49 PM To: sniffer@SortMonster.com Subject: Re: [sniffer] RAID Levels for Spool Folder IMO, Software RAID is not the way to go on a busy machine. You will save a measurable amount of overhead by going with hardware based RAID of any sort since the controller should handle the processes associated with the RAID. Note that this isn't the case with inexpensive RAID controllers such as the cheaper IDE and SATA controllers which still place a fair burden on the OS/processor. True RAID cards also offer additional cache which can speed up the performance on reads, and also on writes if you are battery backed up (otherwise don't use write caching because you could lose or corrupt data during a power outage). There's also several common misconceptions about what is proper to do for a mail server. RAID 5 is the best choice under almost all conditions. The trick here is that while RAID 10 offers both redundancy in mirroring and speed in striping, most servers have a limited amount of space for disks. So a server with 6 disks will operate with the speed of 3 disks spanned in a RAID 10 configuration, but 6 disks in RAID 5 will operate as 5 disks spanned plus a little bit of overhead, though not nearly enough so that it falls short of the performance of just 3 disks in a simple span. Therefore RAID 5 should be the default choice for speed in such an environment. Another misconception is that data is always striped in RAID 0 or RAID 5. This depends on the file size and the stripe size. Most stripes are 64 KB (configurable in most setups). If you have some form of striping for your spool drive, most messages fall far under 64 KB and will only get written to one disk (CRC will also get written in RAID 5). Therefore for a spool folder, RAID 5 with 3 drives (the minimum), will perform rather closely to RAID 5 with 10 drives since most files will only land on one disk (with the other corresponding stripes containing no data). The MFT however for a drive with a lot of files will grow to be quite large and benefits from having multiple disks, and opening very large files such as logs will also benefit from having many disks. There is also an advantage to seek times when having multiple disks, especially if you keep your partitions sized small for performance. I've run a dual processor 3.06 Ghz server with both 6 Seagate 15,000 RPM drives in RAID 5 and the same with 3 Seagate 10,000 RPM drives in RAID 5 running on a less capable controller, and there was no impact on performance while the server was handling over 125,000 unique messages a day. The only noticeable difference was the time it would take to open a 500 MB log file, or the time it would take to enumerate the file names from the MFT on a partition that contained tens of thousands of files in the root. It seems quite apparent that with modern processors, even in dual processor configurations, that you will run out of CUP cycles long before you run out of disk I/O in a well managed RAID 5, 3 drive configuration on an IMail/Declude/Sniffer server. Take note that the log files for Declude, Sniffer and IMail all become massively fragmented, and if you don't have a process to remove these from active partitions on your server or defragment them, then performance will be severely impacted. I run a job hourly that copies all such logs to a different partition and combines them with older
Re: [sniffer] RAID Levels for Spool Folder
I would just RAID 5 the whole setup. With your 6 drives, you get the read performance of 4 drives on any partition in this setup, plus you have a hot spare, and the write performance of close to 4 drives as well. This is a lot better than your config with a mirrored set of drives and a RAID 5 array that reads and writes at about the capacity of two. You will definitely get more bang for your buck with it all being one RAID 5 array, and you get much better redundancy that way as well. As far as splitting things up, there are only 4 things that I feel need to be separate: 1) OS and Executables 2) Spool 3) Mail Boxes 4) Log File Archive Some may split them up further, but I don't think it would have any real effect. The reasons for splitting things up is primarily for the purpose of fragmentation. Your OS and Executables should be mostly static so long as you don't log to that partition, and it might be more advantageous to have all executables on one partition rather than two or more in terms of seek time. The fact that Windows caches frequently accessed files mitigates the effect here so that you probably won't notice the improvement of having everything all together on one partition, so you can do that to taste if you wish. The Spool will get very heavily fragmented, and it is the best place to log to since IMail can't be made to log elsewhere, by Declude and other applications can be made to log there instead of under the IMail tree. Only Sniffer can't be told to log elsewhere at this time, but thankfully the log is not as heavily fragmented as others due to it's comparatively small size. The Mail Boxes should also go on their own partition because of fragmentation, and also for personal preference of organization. They don't get heavily fragmented, but you also don't want to have the spool on the same partition contributing to the fragmentation. The Log File Archive is just simply better on it's own partition since you want to move the files from one partition to another in order to remove the fragmentation. Sizing the partitions is also very important. The outer edges of the disks will run at twice the read and write rates of the inner edges, and if all of your most frequently accessed files are closest to the outer edge there is less seek time, so having your most active partitions nearest the outer edge will improve performance. You do this by not being wasteful with the partition sizes. I put all applications, including IMail and associated programs, along with the page file and OS on the C drive and this is the first logical drive in the array. I am now sizing the primary partition at 20 GB, though you could probably get away with 10 GB. Take note of the page file's max size and the fact that the temp directory is on this drive and things like zipping and unzipping files make use of the temp directory, so you need some extra space. I make the D drive my Web drive (might not be necessary for your setup) and that is quite small. The E drive is for Mail Boxes and I gave it 5 times the actual space in use at this time for growth. The F drive is the Spool, which is 5 GB on my server, and allows for some play in the event that my log archiving mechanism goes bust without me noticing. Then I also have separate small partitions, G and H, for both viruses and held spam, primarily due to the number of files and the size of the MFT, and you don't want this mixed with things on an active partition IMO. Then I have a Log File Archive partition on I which is essentially all the left over space on the array. Here's a summary: C: OS and Executables - 10 GB to 20 GB D: Web - 1 GB E: Mail Boxes - 5 GB F: Spool + Logging - 5 GB G: Spam - 2 GB H: Viruses - 2 GB I: Log Archive - Everything that is left if you wish. I also of course have other devices for storing software and backups that aren't part of the array. With a RAID 5 + Hot spare config, assuming 36 GB drives, you can pack all of the most active stuff into the outer edges of the disks and get screaming performance. Having a large log file archive nearest the center of the platters will have virtually no impact since they would be rarely accessed. Besides the performance gains, you really must consider the redundancy. The fact that in such a setup, you could lose 1 drive and still run at full speed after a period of automatic rebuilding, and lose 2 drives and run at half speed, which would still be plenty, and it would take 3 drives failing to bring your box down...well, that's the reliability that one should want. Like I said earlier, I've done over 125,000 unique messages a day with a sizable Declude setup, Sniffer, two virus scanners, and some Declude helper apps written in VBScript on a server with 3 x 10,000 RPM drives on a zero-channel Adaptec controller with 48 MB of read cache (no write cache) and it had absolutely no issues. That server also ran MS SMTP as the primary gateway along with VamSoft's ORF doing address validation, so