The exchange 2k3 server has operating system - raid 1 , stores raid 5 and logs are raid 1. I have noticed that when email performance suffers the entire lun containg all the guest VM's suffers. I have started collecting stats from the SAN and I have noticed that even with a light load the lun in question appears to have a queue of 3 (limit should be 3 or less according to HP notes). Unfortunately load has been very light this afternoon. I will see what transpires first thing in the morning when load normally increases.
On Tue, Nov 2, 2010 at 2:34 PM, Richard Stovall <rich...@gmail.com> wrote: > On the VMs, are you running multiple disks in guest-OS-based RAID > configurations underneath the EVA's vRAID5 protection, with all the VHDs on > the same LUN? > > What is the exact config for the problematic Exchange server? > > On Mon, Nov 1, 2010 at 11:25 PM, Mark Milo <markmilo2...@gmail.com> wrote: > >> I don't have stats on the eva performance as yet. What would you suggest >> the configuration should be for the main disk array? Just to clarify the >> disk pool is Vraid5 while the disks for individual guest VM's have typically >> been configured as raid 1 for operating system and log disks and raid 5 for >> data. >> >> >> On Tue, Nov 2, 2010 at 2:04 PM, Brian Desmond <br...@briandesmond.com>wrote: >> >>> *Is there any sort of utilization stats on the EVA? Frankly I wouldn’t >>> expect 21 spindles RAID5 to perform very well at all with a mixed workload >>> like this. * >>> >>> * * >>> >>> *Thanks,* >>> >>> *Brian Desmond* >>> >>> *br...@briandesmond.com* >>> >>> * * >>> >>> *c – 312.731.3132* >>> >>> * * >>> >>> *From:* Mark Milo [mailto:markmilo2...@gmail.com] >>> *Sent:* Monday, November 01, 2010 9:14 PM >>> *To:* NT System Admin Issues >>> *Subject:* Re: Virtual Machine Disk Configuration >>> >>> >>> >>> The san has 21 FC disks configured as a raid 5 array. There have two luns >>> configured from this array. All 15 guest machines are located on one of >>> these luns (including exchange). The performance hit appears mainly on >>> exchange but other servers are suffering as well. The interesting thing is >>> that IOPS for the entire lun holding the 15 VM's is only around 600 at the >>> time that the disk busy is around 95%. The disk queue for that lun is around >>> 4 - I will need to set up longer term monitoring to get the read/write times >>> when the disk set gets busy (which is usually first thing in the morning) >>> >>> On Tue, Nov 2, 2010 at 12:35 PM, Sean Martin <seanmarti...@gmail.com> >>> wrote: >>> >>> We'd need a lot more info. Start with your disk configuration. How many >>> disks make up the storage pool or raid group the exchange lun is allocated >>> from? Are there other luns allocated from the same pool? How many? Which >>> exchange resources reside on the lun (stores, logs, etc.)? >>> >>> >>> >>> Disk busy warnings buy themselves don't hold much weight, unless of >>> course there are noticeable performances issues, which you alluded to. >>> You'll want to capture disk queue and disk latency stats ( avg disk queue >>> length and avg disk sec/reads and writes). Avg queue depth should below X >>> (where X equals the number of physical disks. Read/write times should avg >>> below 20ms and not spike above 50ms. >>> >>> >>> >>> I'd also recommend running the exbpa and the troubleshooting assistant. >>> >>> >>> >>> If you can gather I/O stats you may be able to determine if the disks can >>> handle the load. If it doesn't appear Exchange is stressing your resources, >>> you may need your SAN monitoring tools to determine if total I/O (assuming >>> multiple luns share the same disks) are more than the disks can handle. I'm >>> not familiar with EVAs, but overloading disks in a SAN can lead to cache >>> flushing which could affect performance of the array itself. >>> >>> >>> >>> - Sean >>> >>> >>> >>> >>> On Nov 1, 2010, at 5:08 PM, Mark Milo <markmilo2...@gmail.com> wrote: >>> >>> The host is Windows 2008 R2 running Hyper V. The performance is monitored >>> via UPTIME monitoring software but I see similar high reading using perfmon >>> (%disk time). Don't see any error messages at high utilization but the >>> exchange 2003 server is responding poorly when disk is busy. There are 15 >>> guest machines on this particular host but most of them dont use too much in >>> the way of resources. >>> >>> On Tue, Nov 2, 2010 at 11:51 AM, Richard Stovall <rich...@gmail.com> >>> wrote: >>> >>> On Mon, Nov 1, 2010 at 8:33 PM, Mark Milo <markmilo2...@gmail.com> >>> wrote: >>> >>> Hi - I am having issues with "disk busy' on my virtual server host. The >>> host is a DL 580 connected via fiber to an EVA 4400 SAN. >>> >>> >>> >>> Where do you see this message? On the VM? >>> >>> >>> >>> >>> >>> I understand that guest VM's should have disk layouts as per an >>> equivalent physical machine... >>> >>> >>> >>> I don't really know what you mean by that. Can you elaborate a bit? >>> >>> >>> >>> >>> >>> ...but is it best practice to have a separate LUN per guest VM? >>> >>> >>> >>> Not that I've ever heard of. You'd have a management nightmare on your >>> hands pretty quickly if you had a large number of VMs. >>> >>> >>> >>> >>> >>> If so what should the max utilization of the disk on that LUN be? >>> >>> >>> >>> I don't know of a hard and fast number/rule, but if you're using >>> snapshot-based backup method (such as VCB) you definitely need free space on >>> the LUNs for the snapshots to grow until the backup is finished and the >>> snapshots can be eliminated. >>> >>> >>> >>> What virtualization platform are you running? What version? >>> >>> >>> >>> >>> >>> ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ >>> ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~ >>> >>> --- >>> To manage subscriptions click here: >>> http://lyris.sunbelt-software.com/read/my_forums/ >>> or send an email to listmana...@lyris.sunbeltsoftware.com >>> with the body: unsubscribe ntsysadmin >>> >>> >>> >>> ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ >>> ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~ >>> >>> --- >>> To manage subscriptions click here: >>> http://lyris.sunbelt-software.com/read/my_forums/ >>> or send an email to listmana...@lyris.sunbeltsoftware.com >>> with the body: unsubscribe ntsysadmin >>> >>> >>> >>> ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ >>> ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~ >>> >>> --- >>> To manage subscriptions click here: >>> http://lyris.sunbelt-software.com/read/my_forums/ >>> or send an email to listmana...@lyris.sunbeltsoftware.com >>> with the body: unsubscribe ntsysadmin >>> >>> ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ >>> ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~ >>> >>> --- >>> To manage subscriptions click here: >>> http://lyris.sunbelt-software.com/read/my_forums/ >>> or send an email to listmana...@lyris.sunbeltsoftware.com >>> with the body: unsubscribe ntsysadmin >>> >>> >>> >>> ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ >>> ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~ >>> >>> --- >>> To manage subscriptions click here: >>> http://lyris.sunbelt-software.com/read/my_forums/ >>> or send an email to listmana...@lyris.sunbeltsoftware.com >>> with the body: unsubscribe ntsysadmin >>> >>> ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ >>> ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~ >>> >>> --- >>> To manage subscriptions click here: >>> http://lyris.sunbelt-software.com/read/my_forums/ >>> or send an email to listmana...@lyris.sunbeltsoftware.com >>> with the body: unsubscribe ntsysadmin >>> >> >> ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ >> ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~ >> >> --- >> To manage subscriptions click here: >> http://lyris.sunbelt-software.com/read/my_forums/ >> or send an email to listmana...@lyris.sunbeltsoftware.com >> with the body: unsubscribe ntsysadmin >> > > ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ > ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~ > > --- > To manage subscriptions click here: > http://lyris.sunbelt-software.com/read/my_forums/ > or send an email to listmana...@lyris.sunbeltsoftware.com > with the body: unsubscribe ntsysadmin > ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~ --- To manage subscriptions click here: http://lyris.sunbelt-software.com/read/my_forums/ or send an email to listmana...@lyris.sunbeltsoftware.com with the body: unsubscribe ntsysadmin
