Because the limiting factor is still the interconnects at the array side. Depending on the implementation, its possible to have 10 servers (each capable of ~130MBps) connected through switches to 1 storage cabinet, even with 2 interconnects (therefore ~260MBps). Do the math...
1300MBps potential load against a connection that can only handle 260MBps max throughput. What happens on Ethernet when load > bandwidth? It drops or queues packets. SANs have to do something similar. ------------------------------------------------------ Roger D. Seielstad - MCSE Sr. Systems Administrator Inovis - Formerly Harbinger and Extricity Atlanta, GA > -----Original Message----- > From: Shea, Linchi [mailto:[EMAIL PROTECTED]] > Sent: Wednesday, October 09, 2002 5:26 PM > To: NT 2000 Discussions > Subject: RE: Basic SAN question > > > The issue is that we are not just talking about any network. > The network from the server to HBA, to the swicth fabric, and > to EMC is a pretty fast one. With two active dual HBAs, you > get ~130MBps for > sequential loads, and that the throughput of an older 32bit > 33Mhz PCI bus. I know the newer 66MHz/64bit PCI bus has a > much higher throughput for the same sequential loads. > > From some of tests, I can get better IO performance on a EMC > SAN disk than on a locally-attached disk where the paging > file may reside for almost any type of IO profiles. Hence, it > puzzles me. > > Linchi > > > -----Original Message----- > > From: Ken Coughlin [mailto:[EMAIL PROTECTED]] > > Sent: Tuesday, October 08, 2002 8:34 AM > > To: NT 2000 Discussions > > Subject: RE: Basic SAN question > > > > > > If you think about how much data and how often data is moved > > into and out of > > a paging file, I would be wary of putting any paging file on > > a network too. > > Always opt to put you paging file on a local drive. > > > > Ken > > > > -----Original Message----- > > From: Shea, Linchi [mailto:[EMAIL PROTECTED]] > > Sent: Monday, October 07, 2002 5:36 PM > > To: NT 2000 Discussions > > Subject: RE: Basic SAN question > > > > > > The EMC engineers on site here seem to be very wary of going totally > > diskless, and the answer to why is because the paging files > > when you have a > > few hundreds of hosts hooking to the same EMC SAN > > (Symmetrix 3830 or better). I don't totally understand the > > issue yet. But > > they told me it's mostly to do with performance. > > > > Linchi > > > > > -----Original Message----- > > > From: Wes Owen [mailto:[EMAIL PROTECTED]] > > > Sent: Monday, October 07, 2002 4:15 PM > > > To: NT 2000 Discussions > > > Subject: RE: Basic SAN question > > > > > > > > > We are completely diskless. Not doing so would have made > > > the need for an > > > array card and a pair of drives in the server, which is the > > > costs we were > > > trying to avoid. As I could not say removing the redundancy > > > for the drives > > > so long as the page file was still there as then a single > > > drive failure > > > would have brought the server down. > > > > > > Yes, there are some possible issues with the boot from SAN > > > and the paging > > > file if connectivity is interrupted. While I have had a > > > number of persons > > > tell me this is an issue, I have not had anyone that could > > > tell me what > > > actually happens if the problem occurs. > > > > > > After talking with my EMC engineers and talking to customers, > > > some that are > > > and some that are not, no one that I spoke to has ever > > actually had a > > > problem with the page file being on the SAN. The only caveat > > > EMC would give > > > me is that you may want to shut servers down that have page > > > files on the SAN > > > during microcode upgrades. > > > > > > -----Original Message----- > > > From: Shea, Linchi [mailto:[EMAIL PROTECTED]] > > > Sent: Monday, October 07, 2002 3:06 PM > > > To: NT 2000 Discussions > > > Subject: RE: Basic SAN question > > > > > > > > > Owen, as I understand there are still issues with boot > from the SAN, > > > particularly with paging. IN your configuration, do you place > > > the page files > > > on the SAN or on the locally attached drives? > > > > > > Linchi > > > > > > > -----Original Message----- > > > > From: Wes Owen [mailto:[EMAIL PROTECTED]] > > > > Sent: Friday, October 04, 2002 2:51 PM > > > > To: NT 2000 Discussions > > > > Subject: RE: Basic SAN question > > > > > > > > > > > > I would have to offer up a differing opinion. I did a pretty > > > > extensive cost > > > > analysis and was able to show a break even point by using SAN. > > > > > > > > Two factors entered in. Better utilization of disk capacity > > > > and the boot > > > > from SAN capability eliminated not only the drives, but also > > > > the cost of the > > > > array controller. The other soft costs of backup and > > > > manageability provided > > > > the ROI. > > > > > > > > We had a huge problem with our DBA's and developers insisting > > > > they might > > > > need 100 to 400 GB's of drive space over the course of a > > > > server lease and > > > > then when it came off lease they actually only had 20 or 30 > > > > GB's of data on > > > > the system. This allows me to only give them what they need > > > > at first and > > > > then scale it up as needed with minimal disruption. > > > > > > > > -----Original Message----- > > > > From: Roger Seielstad [mailto:[EMAIL PROTECTED]] > > > > Sent: Friday, October 04, 2002 11:15 AM > > > > To: NT 2000 Discussions > > > > Subject: RE: Basic SAN question > > > > > > > > > > > > IMO, the *right* answer is to not buy a SAN for generalized > > > > storage. At the > > > > current price-per-mb rates of SAN solutions vs. Direct Attached > > > > Storage(DAS), I can waste a LOT of locally attached storage > > > > before I break > > > > even moving to a SAN. > > > > > > > > Don't get me wrong - SAN's have their place. I just don't > > > think most > > > > companies need them. And don't even get me started on > NAS boxes, > > > > either. > > > > > > > > ------------------------------------------------------ > > > > Roger D. Seielstad - MCSE > > > > Sr. Systems Administrator > > > > Inovis - Formerly Harbinger and Extricity > > > > Atlanta, GA > > > > > > > > > > > > > -----Original Message----- > > > > > From: Chris Levis [mailto:[EMAIL PROTECTED]] > > > > > Sent: Friday, October 04, 2002 11:27 AM > > > > > To: NT 2000 Discussions > > > > > Subject: RE: Basic SAN question > > > > > > > > > > > > > > > Thanks for the warning. > > > > > > > > > > I do plan on minimizing the number of LUNs, but my boss > > asked the > > > > > question and I wanted to be sure to have the /right/ > > > answer instead > > > > > of the /right-now/ answer. > > > > > > > > > > > -----Original Message----- > > > > > > From: Roger Seielstad [mailto:[EMAIL PROTECTED]] > > > > > > Sent: Friday, October 04, 2002 7:51 AM > > > > > > To: NT 2000 Discussions > > > > > > Subject: RE: Basic SAN question > > > > > > > > > > > > > > > > > > Chris, > > > > > > > > > > > > Most vendors will allow you to slice and dice a SAN > > > array into as > > > > > > many LUNs of whatever size you want. Its absolutely > the wrong > > > > > > thing to do, but it certainly can be done. > > > > > > > > > > > > Any time a phisical platter is partitioned, you're > > > going to take a > > > > > > performance hit - simply put, the heads can't be in two > > > places at > > > > > > once, so if two systems are trying to access data which is > > > > > > physically on the same platter, but logically on > > > different LUNs, > > > > > > there is head contention, and one of the two must > > wait for the > > > > > > other to finish "using" the heads, and then pay the > > additional > > > > > > price of a head seek across the platter to its > > assigned set of > > > > > > cylinders. > > > > > > > > > > > > In the case of your single 500GB RAID5 set in your SAN > > > being split > > > > > > into 300/100/50/50, you have in reality created 4 > > partitions on > > > > > > each spindle, with 60%/20%/10%/10% split on each > > > spindle. With a > > > > > > large number of platters, and larger stripe sizes, its > > > > > > theoretically possible to reduce the chances of > > > contention within > > > > > > the SAN, but realistically speaking, chances are there > > > is going to > > > > > > be some contention, and therefore some performance hits > > > associated > > > > > > with managing your disks this way. > > > > > > > > > > > > Its one of the lies^H^H^H^H omissions commonly done in > > > the sales > > > > > > pitches of the big storage vendors. > > > > > > > > > > > > ------------------------------------------------------ > > > > > > Roger D. Seielstad - MCSE > > > > > > Sr. Systems Administrator > > > > > > Inovis - Formerly Harbinger and Extricity > > > > > > Atlanta, GA > > > > > > > > > > > > > > > > > > > -----Original Message----- > > > > > > > From: Chris Levis [mailto:[EMAIL PROTECTED]] > > > > > > > Sent: Thursday, October 03, 2002 2:07 PM > > > > > > > To: NT 2000 Discussions > > > > > > > Subject: Basic SAN question > > > > > > > > > > > > > > > > > > > > > If you have a RAID-5 array of (let's say) 500GB, can > > > you create > > > > > > > LUNs of an arbitrary size to be presented to the > > > > servers? E.g, a > > > > > > > 300GB, a 100GB, and > > > > > > > two 50GB? Or is there a convention that all LUNs > > have to be > > > > > > > a uniform > > > > > > > size? > > > > > > > > > > > > > > Thanks! > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > ___________________________ > > > > > > > Chris Levis > > > > > > > Applied Geographics, Inc. > > > > > > > > > > > > > > ------ > > > > > > > You are subscribed as [EMAIL PROTECTED] > > > > > > > Archives: http://www.swynk.com/sitesearch/search.asp > > > > > > > To unsubscribe send a blank email to %%email.unsub%% > > > > > > > > > > > > > > > > > > > ------ > > > > > > You are subscribed as [EMAIL PROTECTED] > > > > > > Archives: http://www.swynk.com/sitesearch/search.asp > > > > > > To unsubscribe send a blank email to %%email.unsub%% > > > > > > > > > > > > > > > > ------ > > > > > You are subscribed as [EMAIL PROTECTED] > > > > > Archives: http://www.swynk.com/sitesearch/search.asp > > > > > To unsubscribe send a blank email to %%email.unsub%% > > > > > > > > > > > > > ------ > > > > You are subscribed as [EMAIL PROTECTED] > > > > Archives: http://www.swynk.com/sitesearch/search.asp > > > > To unsubscribe send a blank email to %%email.unsub%% > > > > > > > > > > > > This e-mail and any files transmitted with it are > > > > confidential and are intended solely for the use of the > > > > individual or entity to whom they are addressed. If you are > > > > NOT the intended recipient or the person responsible for > > > > delivering the e-mail to the intended recipient, be advised > > > > that you have received this e-mail in error and that any use, > > > > dissemination, forwarding, printing, or copying of this > > > > e-mail is strictly prohibited. > > > > > > > > > > > > ------ > > > > You are subscribed as [EMAIL PROTECTED] > > > > Archives: http://www.swynk.com/sitesearch/search.asp > > > > To unsubscribe send a blank email to %%email.unsub%% > > > > > > > > > > > > > ------ > > > You are subscribed as [EMAIL PROTECTED] > > > Archives: http://www.swynk.com/sitesearch/search.asp > > > To unsubscribe send a blank email to %%email.unsub%% > > > > > > ------ > > > You are subscribed as [EMAIL PROTECTED] > > > Archives: http://www.swynk.com/sitesearch/search.asp > > > To unsubscribe send a blank email to %%email.unsub%% > > > > > > > > > ------ > > You are subscribed as [EMAIL PROTECTED] > > Archives: http://www.swynk.com/sitesearch/search.asp > > To unsubscribe send a blank email to %%email.unsub%% > > > > ------ > > You are subscribed as [EMAIL PROTECTED] > > Archives: http://www.swynk.com/sitesearch/search.asp > > To unsubscribe send a blank email to %%email.unsub%% > > > > > ------ > You are subscribed as [EMAIL PROTECTED] > Archives: http://www.swynk.com/sitesearch/search.asp > To unsubscribe send a blank email to %%email.unsub%% > ------ You are subscribed as [email protected] Archives: http://www.swynk.com/sitesearch/search.asp To unsubscribe send a blank email to [EMAIL PROTECTED]
