1 INFO-VAX	Sat, 28 Aug 2004	Volume 2004 : Issue 477       Contents:K Re: SKHPC's Latest Take on IPF and Post-Superdome Technology From HP  World  Re: Whither RAID?  Re: Whither RAID? / Re: [TCPIP] How to configure SMTP outgoing only   F ----------------------------------------------------------------------  % Date: Sat, 28 Aug 2004 05:48:29 -0400 * From: "Bill Todd" <billtodd@metrocast.net>T Subject: Re: SKHPC's Latest Take on IPF and Post-Superdome Technology From HP  World= Message-ID: <4Y-dnWcpWsc_ya3cRVn-vA@metrocastcablevision.com>   J Well, it looks like the wild-assed guess I made about large-system Opteron? architecture a couple of days ago was fairly close to the mark:   L http://www.cbronline.com/article_news.asp?guid=2B3F952C-AC0D-4CCB-A285-D9E87 E37A90B   L Four-processor boards using the regular glueless mechanism, with up to eightG such boards joined into a 'ring' (at least it says a ring rather than a J switch; whether it's an HT-based ring is not stated) using directory-basedB cache-coherence and 64 MB of L3 off-chip cache per board to reduce inter-board references.   I If its ring is anywhere nearly as good as POWER5's (which also ties up to J eight boards - MCMs - together), it should scream.  Projected availability about a year from now.   - bill  5 "Bill Todd" <billtodd@metrocast.net> wrote in message 7 news:HdidnXGQnNGT3LDcRVn-ow@metrocastcablevision.com...    ...      The obvious (at least to me,J > though my understanding of the details is too superficial for this to beF > more than a wild-assed guess) solution is to keep the on-chip memoryL > support, and quite possibly the on-chip routing (to allow small groups of,C > say, 2 or 4 processors with their associated RAM), and then use a K > hypertransport switch to connect those small groups together.  This would F > look rather like the POWER architecture (save that POWER requires noI > external HT switch - inter-group routing also occurs on-chip) or even a L > *little* like Superdome (save that Superdome - whether Itanic or PA-RISC -K > lacks on-chip memory support and on-chip routing within the small group).  > H > The one gotcha in the above is that the HT switch must have additionalJ > intelligence in it to support (probably directory-based) cache coherenceE > *between* the small groups, because the on-chip coherence mechanism  doesn't K > scale beyond 8 processors (and 8 may in fact stretch it a bit - we should H > know soon, as 8-processor Opteron systems are starting to appear).  If thatB > is feasible, then a larger Opteron system should exhibit scaling< > characteristics closer to EV7 and POWER than to Superdome.   ...     the fact that it J > can trounce Itanic in small system configurations with only 1/6 as large anJ > on-chip cache is impressive in its own right, but scaling it up to largeJ > systems will require either superb interconnects (possibly of the form IJ > described above) or a larger cache to cut down on inter-group referencesE > (while EV7 gets by on pure interconnect speed, POWER does use large F > auxiliary off-chip caches to help here, and the hypothetical OpteronB > inter-group switch could provide them for Opteron if necessary).   ------------------------------  % Date: Sat, 28 Aug 2004 10:29:02 +0200 * From: Paul Sture <nospam@sture.homeip.net> Subject: Re: Whither RAID?* Message-ID: <2pau2eFj2ickU1@uni-berlin.de>   Andrew Harrison wrote: > Paul Sture wrote:  >  > : >> In that conext, I found this comment interesting. From: >> >> >> >> "The Horrors of RAID 5  >>E >> Using RAID 5 for a high-update Oracle system can be disastrous to  J >> performance, yet many disk vendors continue to push RAID 5 as a viable ( >> solution for highly updated systems." >>1 >> http://www.dba-oracle.com/art_dbazine_disk.htm  >> > B > The operational phrase used here is "can be". RAID 5 performance? > can be bad but that depends entirely on the efficiency of the 6 > RAID 5 implimentation used by the storage subsystem. > > > HDS 99XXX arrays have a very efficient RAID 5 implimentation> > so efficient in fact that very few systems ever use RAID 0+1  > for improved DBMS performance. > = > On the other hand EMC Symetrix Arrays have something called < > RAID-S which is EMC's attempt at something like RAID 5 andA > it performs terribly for random writes and hence is very seldom @ > used for DBMS's in fact EMC's techies move heaven and earth toA > persuade you not to use it for anything that looks like a DBMS.  >   ! That puts it into context thanks.    ------------------------------  % Date: Sat, 28 Aug 2004 10:15:00 +0100 < From: "Alex Daniels" <AlexNOSPAMTHANKSDaniels@themail.co.uk> Subject: Re: Whither RAID?5 Message-ID: <41304d22$0$6158$db0fefd9@news.zen.co.uk>   9 "Rob Young" <young_r@encompasserve.org> wrote in message  - news:LYCrlR1WkHkK@eisner.encompasserve.org... 8 > In article <2p8lk9FieqdmU1@uni-berlin.de>, Paul Sture # > <nospam@sture.homeip.net> writes:  >> Dr. Dweeb wrote:  > I >>> It is a common mistake to think that RAID-X is some kind of panacea.  
 >>> Only aF >>> correct and regular backup strategy can save you from the sort of 
 >>> failure I I >>> experienced.  Indeed, it may be exactly this type of failure that the  >>> article referred to. >>>  >>: >> In that conext, I found this comment interesting. From: >> >> >> >> "The Horrors of RAID 5  >>D >> Using RAID 5 for a high-update Oracle system can be disastrous toI >> performance, yet many disk vendors continue to push RAID 5 as a viable ( >> solution for highly updated systems." >>1 >> http://www.dba-oracle.com/art_dbazine_disk.htm  >> > < > That's because he is trapped in the past or thinking about0 > vendors that had broken RAID5 implementations. > = > He talks about IO contention and file placement, blah blah. ; > But he is missing the changes that have taken place in IO : > subsystems.  Imagine the shock and horror of having your> > log files and redo files share physical spindles!  Of course= > if you are spread across 100+ spindles (EVA) - no big deal.  > C > Or IBM's SAN Volume Controller or many other levels or techniques  > of virtualization. >  > ? > But that would require a change in his thinking so instead of ? > talking about IO throughput or IO latencies being a bad thing C > it is best to "keep it simple" and talk about RAID5 being "bad" - < > but it isn't.  A RAID5 on a modern storage subsystem is in= > most cases a good thing.  This statement of his is an over-  > generalization:  > J > "In sum, it is clearly problematic for any company with high volumes of 	 > updates L > to use RAID 5, and Oracle10g with Automatic Storage Management (ASM) wants$ > Oracle customers to use RAID 1+0." > D > "High volumes of updates."  Okay - define your criteria.  Vacuous. > 7 > Shoot - even EMC with DMX does RAID5 right - finally.  > ( > Finally, this one is an absolute hoot: > I > "Solid state disk is getting cheaper, and may soon replace traditional   > diskI > devices. Many Oracle customers are using solid state disk for high I/O   > data, > files such as TEMP, UNDO, and REDO files." > F > Has he ever priced these?  Last I looked (over a year ago) I found aD > 4 Gbyte one for $64000, I just did a quick Google and found an old' > 536 MBbyte Digital EZ one for $14000.  >  > Rob     K Well if people are paying that, Iv got a bunch of Digital EZ ones and I'll  + let them go for an order of magnitude less.    Alex     ------------------------------  + Date: Sat, 28 Aug 2004 11:07:21 +0000 (UTC) 6 From: peter@langstoeger.at (Peter 'EPLAN' LANGSTOEGER)8 Subject: Re: [TCPIP] How to configure SMTP outgoing only1 Message-ID: <newscache$ial53i$hk31$1@news.sil.at>   h In article <aus-8E68E6.14380527082004@news.cis.dfn.de>, "Hans M. Aus" <aus@vim.uni-wuerzburg.de> writes:j >In article <newscache$ahgfyh$g2o1$1@news.sil.at>, peter@langstoeger.at (Peter 'EPLAN' LANGSTOEGER) wrote:J >> You may know, that I'm not the expert in TCPIP SMTP (since I use MX and, >> also TCPware) so thought I ask here, too: >>  3 >> Is it possible to run TCPIP SMTP outgoing only ?  >>  1 >> I mean, to don't run a service on TCP port 25, C >> but enable local VMS users/applications to send SMTP mails out ?  >>  K >> On MX, this is easy (because SMTP and SMTP_SERVER are different agents).  > G >I just saw your post. We have TCPIP services set up to send us e-mail  K >when routine tasks are finished in the night. There is no incoming email.   >  >Do you still want details?    I'm too curious to say no ;-)   H 1) Do you send SMTP mails (to the next mailhop, maybe to the big companyJ mailserver with logging, restrictions and antivirus features) with TCPIP'sJ SMTP client or with something different (like NETSCAPE MAIL, MOZILLA MAIL, PERL SMTP Module, NBL, ...) ?   M 2) If you use TCPIP's SMTP for outgoing mail, do you have TCPIP's SMTP server " running on the same machine also ?  K 3a) If you do, how do you manage it with your security department when they B complain about "smtp port open, not neccessary/wanted, close it" ?  K 3b) If you don't, how did you do that and is it some kind of 'fool proof' ?   L I did a TCPIP SET CONFIG SMTP/ACCEPT=HOST=LOCALHOST and still hope that such: a 'half open' port will be enough for the security guys...   --   Peter "EPLAN" LANGSTOEGER % Network and OpenVMS system specialist  E-mail  peter@langstoeger.atF A-1030 VIENNA  AUSTRIA              I'm not a pessimist, I'm a realist   ------------------------------   End of INFO-VAX 2004.477 ************************