Yep, so we have specified a limit of 200Mbps in our service offering.. But again, I think windows workloads would work fine if the infrastructure is designed from the get go for such a use case. Thanks everybody for your feedback, I think I have got my answers.
On Wed, Nov 13, 2013 at 6:15 PM, Conrad Geiger <cgei...@it1solutions.com>wrote: > The additional 5-10 users shouldn't be such an extreme load. > How many Mbps were you using with the 5-10 users? > > I am trying to clarify if the SAN or the storage network is the bottleneck. > > In either case,as previous stated it does all go back to capacity/workload > planning. > > I know this is getting beyond cloudstack, but on the ZFS box you can run > 'zpool iostat -v' to see your IO and throughput averages. > > Be careful on the dedicated ZIL, it can quickly become a bottleneck if you > don't purchase an SSD capably of the load. > > Junaid Shahid <shahid.jun...@gmail.com> wrote: > > > Yeah with 5-10 users only :) > > Also I think we don't have any write-cache (called ZILs in the ZFS lingo, I > think) on the storage server too, so SQL would be even more problematic > there.. > > > On Wed, Nov 13, 2013 at 5:53 PM, Conrad Geiger <cgei...@it1solutions.com > >wrote: > > > Are you really saturating you GigE link with only 5-10 users. > > > > It sounds like you may be running out of IOs, SQL is usually a very write > > intensive workload. > > > > Junaid Shahid <shahid.jun...@gmail.com> wrote: > > > > > > Thanks Todd! > > > > Well I think the service offering is at 200Mbps.. Also I we are not using > > any link aggregation at all. Let me float these ideas to my team. Thanks > > for your feedback! > > > > > > On Wed, Nov 13, 2013 at 5:36 PM, Todd Pigram <t...@toddpigram.com> > wrote: > > > > > Junaid, > > > > > > what did you set the the network rate to in the exchange service > > offering? > > > Depending on your backend network setup for that offering you may get > > > better results with setting it to a '0' for unlimited. On my internal > > CCP, > > > our SQL servers service offering has network rate to '0' as I am using > a > > 4 > > > NIC LACP bond. > > > > > > just food for thought > > > > > > Todd > > > > > > > > > On Wed, Nov 13, 2013 at 6:37 AM, Junaid Shahid < > shahid.jun...@gmail.com > > > >wrote: > > > > > > > Hi all, > > > > We are running a mixture of Windows and Linux VMs under different > > > accounts > > > > on our cloud, that is based on CloudPlatform 3 (I know that it's a > > > mailing > > > > list for ACS, but I still need your feedback so read on please :)). > > > > > > > > The Primary storage is based on iSCSI with GigE link, and Xen > > hyperviser. > > > > > > > > Now the problem is that whenever we run Windows OSes with > applications > > > like > > > > Exchange, Sharepoint and particularly MS Lync (that includes AD and > > MSSQL > > > > as pre-requisites..), the GigE link to Primary Storage becomes so > > > congested > > > > that it affects the whole cloud environment. Nothing remains usable > > > > anymore, the performance of Linux VMs also is affected in the > process. > > > > > > > > So what does your experience say, what should we do: > > > > 1) Segregate the Windows VMs to their own cluster and their own > > separate > > > > Primary storage. > > > > 2) Use local storage for the "pre-cloud era" traditional Windows > > > workloads > > > > such as MS Exchange etc. > > > > 3) Is cloud environment feasible at all for Hosted Exchange and the > > > like, > > > > as Local storage that runs on the speed of the motherboard > back-plane, > > of > > > > course cannot be matched by a GigE link alone. > > > > > > > > Awaiting your valuable feedback all :) > > > > > > > > -- > > > > Regards, > > > > Junaid Shahid, > > > > TODO:______ > > > > > > > > > > > > > > > -- > > Regards, > > Junaid Shahid, > > TODO:______ > > > > > > > -- > Regards, > Junaid Shahid, > TODO:______ > > -- Regards, Junaid Shahid, TODO:______
