--------------------------------------------------------------------- NOV-MEM3.DOC -- 19980320 -- Email thread on NetWare memory management --------------------------------------------------------------------- Feel free to add or edit this document and then email it back to faq@jelyon.com Date: Tue, 4 Feb 1997 15:42:33 -0600 From: Joe Doupnik Subject: Re: Register Memory >I replaced a 486/33 with 16MB for a motherboard with an Intel P133 >motherboard and 32MB. >I turned auto register memory off and set the > register memory 1000000 1000000 Ok, that's the first mistake. It is most often not needed on PCI bus machines. It's origin is helping ISA bus machines. But you plowed ahead anyway and probably ran out of usable memory below 16MB which in turn corrupted the disk i/o material. May I strongly recommend people approach this item carefully. Try the machine without startup.ncf and without autoexec.ncf. See what MEMORY says on the console. If EISA bus one MUST configure EISA via its setup program shipped on a floppy with your machine. PCI machines often need no help, but some still not report clearly and thus REQUIRE the NW patches to function properly. In such "Require" cases do not fill up the disks and then think about patches. Instead install a skeleton NW, patch it, and only after MEMORY shows proper values proceed to add user files. Some PCI bus machines not reporting properly may also benfit from loading HIMEM.SYS (and only that) in config.sys. I view such machines as not suitable as servers, but then I don't have the problem. If one pushes ahead with a misconfigured server then as noted in the first paragraph all heck can break loose down in the code, with nasty consequences. Finally, we continuously recommend that folks read the interesting material in the list's FAQ before coming to us with difficulties. It saves wear and tear on everyone, and it's educational. Joe D. ------------------------------ Date: Fri, 14 Mar 1997 23:41:53 EST From: seanstanton@juno.com (Sean M Stanton) To: netw4-l@ecnet.net Subject: Re: HELP !!!! >Mounting Volume SYS2 >Insufficient memory available to mount volume. >Volume SYS2 not mounted The reason this happens is that when NW3.11 (later versions also, I assume) mounts it's first volume after starting up, it allocates memory starting at the top of available memory for it's FAT and directory table caching areas. It will then only allocate memory below that address for cachng FAT's and directory tables on subsequent volume mounts. What you need to do is, as Bob mentions below, is to move the AUTOEXEC.NCF file to your DOS boot drive where your SERVER.EXE and STARTUP.NCF are. But you also have to move the LOAD xxx.DSK commands from your STARTUP.NCF file to your ATOEXEC.NCF. If you don't do this, SYS: will still auto mount at the end of the STARTUP.NCF and you won't avoid the problem. Then put your REGISTER MEMORY command as the third line in the AUTOEXEC.NCF, after the FILE SERVER NAME=xxx and INTERNAL IPX NETWORK=yyyyyyy lines, but before before you actully load the .DSK drivers. Then, after the LOAD xxx.DSK lines in the AUTOEXEC.NCF, you need to add a MOUNT ALL command. So the top of your AUTOEXEC.NCF on the C: drive would look like: File Server Name=myserver IPX Internal Network=123456 Register Memory 1000000 7000000 Load Myscsi.dsk Int=x Port=yyy Mount All ------------------------------ Date: Thu, 20 Mar 1997 19:01:12 -0500 From: Larry Hansford Subject: Re: Adding Memory to 4.11 Server >I have a Micron server running IntranetWare 4.11. It came from the >factory with 32MB of RAM and Netware saw all of the RAM with having >to add the Register Memory line to the Startup.ncf. > >I just added 64MBs of additional RAM to bring the RAM count up to >96MB. (the system counts 640K + 97280MB at bootup.) Unfortunately, >Netware still only sees 32MB. I tried adding the following line to >the STARTUP.NCF, but it does not seem to affect the amount of RAM >seen by Netware: > >REGISTER MEMORY 1000000 5000000 > >I also ran the server's EISA configuration program and added the >additional RAM to the EISA settings. > >What else can I try to get the RAM to be recognized? Ensure that you Do Not have a Set Auto Register Memory ON command in your Startup.Ncf file. This is normally the cause of such symptons in that it Auto Register the 32 Mb of RAM, which will cause the Register Memory command to be ignored. You may want also to test the memory configuration by starting the server without loading the Startup.Ncf or Autoexec.Ncf files and then doing a Memory command to see how much memory is seen by the server without you adding any commands at all. ------------------------------ Date: Fri, 21 Mar 1997 10:01:34 +0800 From: Brett Looney Subject: Re: Large volume mounts on netware 3.12 >Have a 9G drive with a 5G volume on a 32M 486 running NW 3.12 with all >the updates in PATLST.txt installed. >The message is : > Insufficient memory to mount volume I'm not sure if you're having the same problem I've seen, but here's my 2c anyway. I'm going to assume that you register memory in the AUTOEXEC.NCF. If not (and the server auto-registers memory) then ignore everything from here on in. So, if you are registering memory, it only gets regestiered after the disk driver loads - and for some reason (architecure wise in the OS) this affects whether you can use all the memory in the server for dir entry tables and the like. So, to fix this you need to register memory _before_ loading the disk drivers - but in NW 3.x you can't register memory in STARTUP.NCF - so we have to load the disk drivers in AUTOEXEC.NCF after registering the memory - but if you do that, you can't get to AUTOEXEC.NCF because it's on SYS: and you can't get there without the disk drivers (catch 22...). So, do this: Take the disk driver commands out of STARTUP and put them in AUTOEXEC. AUTOEXEC.NCF should then look like this: file server name ipx internal network register memory load mount sys And the trick is - put this AUTOEXEC.NCF on the C: in the same directory as the STARTUP.NCF. The drawback is that you can't edit this file using INSTALL.NLM or SYSCON (but you can use EDIT.NLM) and any changes you make to the AUTOEXEC.NCF that is in SYS:SYSTEM don't happen! Hope this helps - this resolves most of our register memory problems with 3.x... ------------------------------ Date: Fri, 21 Mar 1997 10:14:14 +0800 From: Brett Looney Subject: Re: Adding Memory to 4.11 Server >I just added 64MBs of additional RAM to bring the RAM count up to >96MB. (the system counts 640K + 97280MB at bootup.) Unfortunately, >Netware still only sees 32MB. I tried adding the following line to >the STARTUP.NCF, but it does not seem to affect the amount of RAM >seen by Netware: > >REGISTER MEMORY 1000000 5000000 Long story here - it has a point that hopefully will shine some light... I noticed a strange thing with some boxes. When you boot and go into protected mode, the BIOS gets relocated to somewhere other than at the end of the first 16Mb. In some boxes (I think yours) it goes to the end of the first 32 Mb, in others (like mine) it goes to the end of the first 64Mb. What this results in is a gap in the memory map which the REGISTER MEMORY command can't deal with. In effect, you need to put in two register memory commands - one to get to the gap, and one that starts 384k above the gap and goes to the end of RAM. So, on my machine with 96Mb of RAM with the gap at 64Mb I do this: (my first 64Mb gets auto-registered) REGISTER MEMORY 40f0000 1f10000 On your machine (guessing here - it took some trial and error to get mine right), it might be: REGISTER MEMORY 1000000 1000000 (the second 16Mb) REGISTER MEMORY 20f0000 3f10000 (the next 64Mb, almost) How did I figure there was a gap? I couldn't register memory starting at the 64Mb boundary! Some time working with a hex calculator may also help... ------------------------------ Date: Sat, 22 Mar 1997 12:13:55 GMT From: Teo Kirkinen Subject: Re: NCP Requests & 100% Utilization >LRU sitting times run from 40 seconds to 2 minutes during these times. >Cache buffers were at roughly 78%. My compression start and end times >are in the 3-6am range. We're running a Compaq 1500R with 11GB and >160MB of RAM -- compression and BSA on all but SYS volume (600MB) and >no additional name space (yet). I would add more memory to the server until the LRU sitting time stays above 15 minutes. This has made our servers more responsive during peak work hours. The magic number comes from some NDS update interval. I learned it at last year's Brainshare Europe. ------------------------------ Date: Tue, 1 Apr 1997 20:11:48 -0600 From: Joe Doupnik Subject: Re: "...out of available memory" >One of my 4.10 servers, running on a PCI bus with Adaptec's 2940 adapter, >64MB Ram and 2x4GB disks (Fujitsu & Quantum), has abended, and following a >restart came up with the message "Cache memory allocator out of available >memory". > >The NCF files are both on c:\nwserver, memory is registered correctly (I >assume, as the server memory command reports it all and it did work >without complaint over the last 40 days). Here are the files: Ok, may I correct the logic. The server is likely in bad shape, and the error message really does mean what it says. Why? Look at what you've done below. You turned off the ability for NW to sense memory about 16MB. There must have been a reason, and that is normally a large Red Flag that things are amiss. Then you do a REGISTER MEMORY command hoping to recover smoothly (and it doesn't). And you load the disk driver within autoexec.ncf because clearly it did not work when loaded properly in startup.ncf. My, what amount of warning is necessary? If that is an EISA bus machine then a) you must, positively must run its EISA configuration program and set the memory quantity exactly, and b) also read the FAQ where this situation has been beaten to death. Never use REGISTER MEMORY with EISA or MCA buses. If it is not an EISA bus machine then please get rid of it and acquire one with a working motherboard. Before you junk it try loading himem.sys from DOS; just that and nothing else in config.sys. If it works then still get rid of the motherboard but this gives you time to wait for a replacment in some comfort. As things stand now all your disk structures load at 16MB and work down, and they collide with the main o/s. That is fatal. Joe D. >**** c:\nwserver\startup.ncf **** >LOAD C:\NWSERVER\PATCHES\410PT6\NATIVE\PM410.NLM (a little behind) >PMLOAD C:\NWSERVER\PATCHES\410PT6\NATIVE\START_PT > >SET MINIMUM PACKET RECEIVE BUFFERS = 200 >SET AUTO REGISTER MEMORY ABOVE 16 MEGABYTES = OFF >SET RESERVED BUFFERS BELOW 16 MEG = 60 > >**** c:\nwserver\autoexec.ncf **** >file server name HUJI_MDSTUD >ipx internal net 8440A10A >register memory 1000000 3000000 >load aic7870 slot=29 (latest from Adaptec) >mount sys >mount usr >mount apps >sys:system\autoexec.ncf ------------------------------ Date: Tue, 22 Apr 1997 16:47:57 CST From: John Delano Subject: Re: Paradox 4.5 and NW 4.1 >The sequence "login, run paradox, optionally work with paradox, >exit paradox, logout, login, run paradox" causes 'memory corruption >error, cannot load command, system halted' message. I found a >document on support.novell.com "discussing" this issue which says, >that solution is 'Customer run MEMMAKER that solved the problem.' > >Well this makes almost no sense. I found that the only solutions of >this problem are: >a) do not load EMM386 in config.sys >b) load EMM386 with EMS support (even if EMS is not needed) > >The workstation is using VLM client. The server has all current >patches installed. The list of actions that _do not_ solve the >problem follows: >loading all drivers low while EMM is present, upgrading to the >newest VLM client, loading connection table low, excluding >unnecessary resident programs from autoexec, playing with 'load low >... = on/off" in net.cfg. > >QUESTION: Is there any idea what can cause this problem and how to >solve it having the maximum UMB memory available ? We had a very similar problem on our system when we were using VLM's on the client. We would get the "Cannot load command.com system halted" error, but more frustrating was the fact that it did not happen *every* time we ran paradox. We discovered that the problem was directly related to Emm386, and by using the NOEMS option we were able to reduce the occurrence of the error. However, after using this option, we were no longer able to run Windows after exiting from paradox without first rebooting the PC (system would hang, etc.). Our solution was to upgrade the clients to Client32 and remove the EMM386 option alltogether, since client32 doesn't take up nearly as much conventional memory as VLM's. Since we are using 4.11 for our internet gateway we needed to upgrade anyway, but it would still be nice to use EMM386 to free up even more room. One thing we didn't try was a third party expanded memory manager like QEMM, etc. Might be worth a shot??? ------------------------------ From ???@??? Wed Jan 08 13:07:43 1997 Date: Wed, 8 Jan 1997 09:30:53 -0500 From: Dave Henderson Subject: Re: CACHE MEMORY ON 3.11 SERVER. I had this problem with my NW 3.12 server (a PCI motherboard) and received some good advice from the list. The solution that worked for me was as follows: Try Joe's acid test below. Start the server using "server -ns -na" and then at the server prompt run the "memory" command and see if NetWare finds all your memory. If it doesn't, downloaded the NW 3.11 patches from one of the netlab sites( I believe I seen a patch file there for 3.11 . One of the patches that was in the 312PTA.EXE download is LSWAP.EXE or LSWAP.NLM that runs LOADER.EXE. It is a static patch which means you run it only once. I ran it and it cured my problems. Beware, when I loaded LSWAP.NLM it didn't take the first time. It was looking to \NWSERVER directory for certain files (LOADER.EXE and SERVER.EXE I believe) and \NWSERVER was not there of course. I suspect this NLM was developed in NW 4.x (therefore \NWSERVER directory) and then modified or whatever to be used in NW 3.12 and they failed to change the directory reference. I had to xcopy/s/e my \server.312 to \nwserver, start the server from the \nwserver directory, load the nlm and then down the server and xcopy/s/e \nwserver back to \server.312. >Joe wrote: >> Alas, the acid test is a good one. You didn't mention which version >>of NetWare is involved, so let us assume it is NW 4.10. In that case there >>is help available to NW in decoding what you system has in memory capacity >>even though the cpu and CMOS memory decline to reveal that information >>conveniently. PCI bus machines can be that difficult. The help is installation >>of the compendium patch (translation: "really big") 410PT6.EXE, located in >>updates\nwos\nw410 together with all the other things needed adding, and >>available from ftp.novell.com and faithful mirrors thereof. > > To quote from that patch: >> >>LOADER EXE >>============ >> >> SYMPTOM: >> >> 3) ADDED support for using the new BIOS call int 15 sub >> function E8 for memory detection on PCI machines and other >> ISA machines that have more that 64 Meg of memory. >> >> 13) ADDED the new DoProtectedModeBIOSCall to the loader for >> use by ODI with new PCI device drivers. This API allows a >> driver to make protected mode calls to the BIOS. >> >> If that still does not help then I have only one more recommendation: >>beat up on the machine's Bios Setup (and any supplemental configuration >>programs shipped with the machine) to set the memory capacity correctly. >>Actually, I have another: try loading himem.sys (not emm386 et al, just >>himem.sys) in config.sys. That can sometimes coerce the machine into revealing >>its true capacity, and other times it just becomes a hazard best avoided. >> And add a guess to the list: turn off all Plug & Play support. That >>stuff easily confuses hardware when poked by the Bios. >> Were I in your shoes I'd strip the machine to bare metal, get it >>working with NW, and only then start adding more peripherial boards. And >>I trust not memory SIMMs, so look around for others to borrow to sort out >>system design difficulties. >> The moral of the story is don't let the machine win without a >>struggle. Apply all the patches and probe every nook and cranny to find >>a way to the goal. Even subtrafuge (himem.sys) is legit. >> Joe D. ------------------------------ Date: Wed, 29 Oct 1997 19:29:22 -0500 From: Larry Hansford Subject: Re: Mirroring drives >I have a NW 3.12 server with a 4 GB HD in it and 64 MB RAM. I want to >add another 4GB HD and mirror the two. > >Do I need to add more RAM because of the second drive? Is anything on >the second drive cached in RAM? No, the mirroring does not require addtitional memory. ------------------------------ Date: Wed, 5 Nov 1997 14:13:46 -0600 From: Joe Doupnik Subject: Re: Loads of Memory >Have you noticed that lately a lot of the suggestions regarding disks, >servers, programs etc seem to have one thing in common ? > >"Loads of RAM", "Gobs of RAM" (That last one sounds like something >slightly gooey actually....), "Tons of memory". > >Are we all that skeptical of Novell that we seem to have this view >(myself included), that to get something to work the best way, simply >make sure it has a zillion MB'ts of memory. > >I have been imagening some of the smaller systems, A Pentium 166 server, >a 1.5Gb disk with a SCSI adapter.......and 512MB RAM to be safe. > >What a wonderfull thought. -------- And it is a productive thought. A few years ago I dug into the matter and launched quite an instructive thread on figuring out fact from urban legend (both popular and enshrined). The result was urban legend was far beyond reason, and in fact had no defensible reasons to offer. The list's FAQ has excerpts of this discussion. Being an engineer and science person I often do not accept opinions as truth, particularly when I can test things easily myself. I strongly, warmly, encourage testing and thinking about the findings. In the case of a nice NW 3.12 server, 64MB can support oodles of disk farm with little difficulty. I have 9GB on one (netlab2) and it has memory to burn. What I do is figure out how much memory is statically consumed by the o/s and peripherals, how much is needed by transitory NLMs such as tape backup programs, and finally how much needs to be kept as scratch cache buffering for files and directories. We easily see that adding disks consumes static memory space, and adding ACTIVE users consumes the dynamic cache stuff. I run a variety of servers, and I'm pleased to say even experienced on-lookers were amazed when I blithely added 4GB to a INW 4.11 server and the free memory figure barely budged. That is what large disk allocation units buy us, and we pay back a fee to acquire subblock allocation. NW 3 has the same disk allocation unit choices, but does not have subblock allocation. Some thought should yield a good allocation unit size for particular NW 3 installations, each installation being different. 4KB is often unnecessarily small. Finally, I get quite concerned about the motherboard being able to cope with memory greater than 64MB. Tis a world of cheap badly designed PC hardware at this time, no thanks to Intel for blunders on PCI chip sets. Because we can't predict the hardware behavior we test it instead. Brand names don't make the hardware run better, and often it runs a lot worse because of cute "embellishments." Again, we gamble some money to get hands on the components to test and observe. So, before heeding advice to throw memory at problems, please read the FAQ, read MONITOR, do some careful thinking and testing, and finally draw conclusions of merit. Finally, to look ahead not very far, we will be running NW servers with virtual memory and with file systems which require very little memory. These are presently known by development code words MOAB and NSS, resp. Joe D. --------- Date: Thu, 6 Nov 1997 09:07:19 +0100 From: "A.H. Bos" Subject: Re: RAM I agree, but this only goes for small sites. When you have less than 50 users at a time and lets say less than 4 GB of Netware volumes an 486 with 32 MB should be enough. We have 2500 users and 0.12 TB (1 Tera Byte = 1024 Giga Byte) of netware volumes. We have 3 servers with 512 MB but this is not enough, now we are building another 3 servers with 512 MB and hope performance will be better for a while. The amount of memory you need is related to the number of users (at the same time), the total amount of volume space and the kind of use of the server (For example: An application server will need lots of memory). ------------------------------ Date: Fri, 7 Nov 1997 11:02:51 +0200 From: Mike Glassman - Admin Subject: Registering memory >>I have Dell PowerEdge SPS2-75 server computer with 64 MB RAM >>installed. It works under Novell NetWare 4.1. >>Originally we bought this comp with 32 MB RAM. After upgrading to 64 >>my NetWare was unable to autoregister memory above 32 MB, so I had to >>write manually REGISTER MEMORY 2000000 2000000. >> >>Recently I've got Internet access and decided to download latest >>patches for NW 4.1 because Novell recommended that. >> >>Yesterday I've downloaded 410PT7.EXE. After installing and reboot I >>discovered that half of my NLMs weren't loaded or loaded but weren't >>work correctly due to lack of memory lack. >>After reboot and loading again I've noticed a message: "memory >>wasn't added". So, it seems that new, patched NetWare is unable >>to register this upper memory at all. >>When I restored original SERVER.EXE everything is went ok, but who >>knows - may be for correct work of all patch NLMs *NEW SERVER.EXE* >>needed ? > >The reason is on some EISA machines NetWare couldn't exactly determine >how much memory is installed. > >With the native Server.exe and Auto Register Memory Above 16 Megabytes >set to On, NetWare found 32 Mb without assistance and another 32 Mb I >specified manually with REGISTER MEMORY 2000000 2000000. > >When I've updated server with 410pt7 REGSITER MEMORY 2000000 2000000 >stopped work. System says that this memory is already registered, but >really available memory was only 16 MB. I realized that I should set >Auto Register Memory Above 16 Megabytes to off and specify REGISTER >MEMORY 1000000 3000000. > >And now everything is works fine. On 90% of EISA machines (not including some IBM PC-Servers such as the 330) In order for the system (Novell etc) to recognize the memory correctly, you need to run the EISA Config program and tell the Mother-Board that you have whatever amount of memory. If you don't do that, or if your Board doesn't support the "telling how much memory option" You might then have a problem with memory being recognised by Netware. As a standard, on EISA machines, you should NOT have to type "Register Memory" at all. If you do, look to your Configuration and see if perhaps you or your tech forgot to define the memory in the configuration utility. I have seen multiple occasions where techs installed aditional memory in EISa machines, and thought that since the memory is counted at boot time, that they didn't have to make any changes in the Configuration of the EISA board. This is totaly incorrect, and can lead to problems later on down the road. ------------------------------ Date: Thu, 27 Nov 1997 15:45:38 -0500 From: Jim Pretty Subject: Server hangs after adding drive Check out the memory calculation utility at http://www.oncss.com/dwilliams It gives options for Intranetware flavours but should be just as valid for 3.11. The biggest factors which determine needed memory are block size, disk size and loaded NLMS. (Don't sweat the last one; if you don't know how much your NLMs are going to need use at least 8MB and preferrably 16MB to be realistic. ------------------------------ Date: Thu, 11 Dec 1997 20:46:09 -0600 From: Joe Doupnik Subject: Re: speed up cheaply and easily? >The note about the cost [of memory] was also interesting, I didn't >know you could get so much memory for so little! -------- Novell has a well written white paper on NW 3/4 memory usage and requirements, in line with the exploration, inference, and deductions we did here live on the list some time ago (see the list's FAQ for the blow by blow story). It is in their Appnotes, which are now on the net via http://developer.novell.com/appnotes/appnotes.htm. The regular web page will get you there eventually. I think this document and a number of others should be read by serious system managers. Joe D. ------------------------------ Date: Fri, 12 Dec 1997 12:59:35 -0600 From: Joe Doupnik Subject: Re: Novell White Paper >The White paper I believe Joe is referring to is in the March 97 >issue. If it's not *the* one, it is a paper on memory. >http://developer.novell.com/research/appnotes/1997/a9703.htm ------- Correct date. The paper title is "Optimizing IntranetWare Server Memory" by Ron Lee. March 1997, 31 pages. Reading all the issues is a good thing, when you find snippets of time to do so. As normal, take issue with what you read until you feel comfortable with the results: just because it is written does not make it... Thanks Tim, Joe D. ------------------------------ Date: Mon, 29 Dec 1997 06:18:01 -0700 From: Hansang Bae Subject: Re: memory problem >> I have a prioris xl 5100 server vith novel ver. 4.1 with the >> patches install .However novell wont recognize more than 64 Mb. or Ram. [snip: read the FAQ, did the startup.ncf thing, not sure if it worked] >There is a bug with 3.12 that requires the register memory command to be >implimented BEFORE the sys volume mounts, otherwise ALL the volumes >mount below the 16 MB range. Even if you have more. This is true for 3.1x but not for 4.1 and above. The reason? You can put the register memory command in the startup.ncf w/o a problem. In the 3.x world, REG MEM could only go in autoexec.ncf... hence the workaround. More of an oversight than a bug, methinks. There are some PCI motherboard that will not auto register memory even with the patches. I would've thought that a Digitial box would be able to register memory. Especially with the patches loaded. I'm not sure how Grondin knew that the DET/FATs were planted at the 16meg level, however. I'm not sure about the digital box, but if it's an EISA box, you *must* run the eisa config utility as NetWare will poll the eisa config setup, *not* the cmos info. ------------------------------ Date: Tue, 6 Jan 1998 11:06:24 -0500 From: Larry Hansford Subject: Re: Memory >I was reading a 3.12 System Admin. book last night and noticed that 3.12 >defaults to 8MB RAM??? I came in this morning and checked MONITOR, >MEMORY UTILIZATION, > >It shows: > >Percent in use 83% >Percent free 17% >Memory blocks in use 32,720 >Memory blocks free 1,160 >Memory bytes in use 7,422,211 >Memory bytes free 1,495,789 > >Is this saying I only have 8MB set for server RAM???? From the server command prompt, type in MEMORY and it will report exactly how much total memory it sees. ------------------------------ Date: Fri, 20 Mar 1998 19:34:02 -0700 From: Joe Doupnik Subject: Re: Need to confirm low memory problem >>NetWare's internal allocation of memory to satisfy the subsequent >>needs of running system threads, as well as the loading and unloading >>of NLMs and their allocation of temporary memory resources, can result >>in significant memory fragmentation and subsequent performance >>degradation. Thus, contrary to popular belief, it can be advantageous >>to down the Server periodically in order to reset the base of >>allocated resources. > >OK, but in general, don't 98% of all the NLM's a server is going to use >get loaded at startup? Of course, for the first day or so there will be >additional memory allocations for user logins and thread processes. >After a relatively short period of time, though, I would expect all that >to steady down and not further fragment the server's memory. > >I have seen servers where NLMs are loaded and unloaded on a daily basis >(so backups can be run, restart a process, etc., etc.). In an >environment like that, you're much more likely to fragment the server's >memory. --------- The first, double marked, comment is correct. Many NLMs make dynamic memory requests of varying size and that leads to fragmentation of available space. The emphasis is on dynamic, meaning allocations are made and released often during normal operation. The Novell web server v3.1, as an example, allocates its own pool for work to avoid just this problem. This is in addition to the perennial memory leak problems present in any complex system. Just how pronounced is fragmentation depends highly on what is in the server and how those items are used. I too think it a generally good idea to reboot the server now and then, but after backups rather than before. Joe D. ------------------------------