Where I work, we are in need of upgrading our ClearCase servers and it's been proposed that we move them into a new (yet-to-be-deployed) VMmare system.
In the past I've not noticed a significant problem with performance with most applications when running in VMs, but given that ClearCase "speed" (i.e. dynamic-view response times) is so latency sensitive I am concerned that this will not be a good idea.
VMWare has numerous white-papers detailing performance related issues based on network traffic patterns that re-inforces my hypothesis, but nothing particularly concrete for this particular use case that I can see.
What I can find are various forum posts online, but which are somewhat dated, e.g.:
ClearCase clients are supported on VMWare, but not for performance issues. I would never put a production server on VM. It will work but will be slower. The more complex the slower it gets. accessing or building from a local snapshot view will be the fastest, building in a remote VM stored dynamic view using clearmake will be painful..... VMWare is best used for test environments
(via http://www.cmcrossroads.com/forums?func=view&catid=31&id=44094&limit=10&start=10)
and:
VMware + ClearCase = works but SLUGGISH!!!!!! (windows)(not for production environment) My company tried to mandate that all new apps or app upgrades needed to be on/moved VMware instances. The VMware instance could not handle the demands of ClearCase. (come to find out that I was sharing a box with a database server)
Will you know what else would be on that box besides ClearCase?
Karl (via http://www.cmcrossroads.com/forums?func=view&id=44094&catid=31)
and:
... are still finding we can't get the performance using dynamic views to below 2.5 times that of a physical machine. Interestingly, speaking to a few people with much VMWare experience and indeed from running builds, we are finding that typically, VMWare doesn't take that much longer for most applications and about 10-20% longer has been quoted.
(via http://www.cmcrossroads.com/forums?func=view&catid=31&id=44094&limit=10&start=10)
Which brings me to the more direct question: Does anyone have any more recent experience with ClearCase servers on VMware (if not any specific, relevant performance advice)?
For ClearCase registry server or license server, why not.
But for ClearCase Vob server or View (storage) server? I think not:
All of our Vob servers are on Solaris10, with zones, and ZFS (for extra large disk capacity).
I have built two VMware clearcase clients (a RH5.3 and a RH4.2). They have 2 GB RAM and a CPU reservation of 2GHz on an ESXi 4.1 hypervisor. IMHO they work ok - faster than the Sun Fire V240/Solaris 10u7 clients I had before. I am thinking of creating both a view and vob servers on ESXi but using raw device mapping in order to speed up things. I do not expect serious performance bottlenecks. As for time sync I solved the issue via VMWare tools installation - no more clock skews since.
I recently have had some experience with VMWare & ClearCase. For one of my environments it was required to use a ClearCase client on a VMWare machine. The specific purpose was to build the code in snapshot views. Previous testing (2yrs ago) had shown that snapshot load times itself were almost 1.5x more on VMWare machine.
However, a recent testing was encouraging and much lag wasnt seen. The config was: Physical Machine was a DL386 G6 with 26GB RAM; VM's had 8GB RAM & 160GB HDD allocated
I've already installed ClearCase VOB servers in VMware on Linux RedHat and Windows.
Check this: http://www.ibm.com/developerworks/rational/library/smart-virtualization-1/index.html
I've been testing a Solaris 10 x86 VM (2.6Ghz physical machine) with 4 cores and 16 Gig of memory against our old V210s 1.3GHz sparc with 16Gig of memory in a NAS environment. The VOB is a copy of production. 1.2Gig database with 9.5Gig source pool. Results so far: - Database load on VM is 2X faster than V210 (10.5min vs 22min) - cleartool find -version lbtype(label) -print ON the server is 2x faster. WARNING BUT: Same command on a client (Ubuntu/Solaris) is about 2X SLOWER.
It appears all the RPC calls from the client to the VM ClearCase server is the issue. I have tcpdumps loaded into wireshark. Nothing stands out. About the same number of network transations, but the VM x86 machine takes longer per call and as a result, the old V210 gets the job done faster when a client is interfaced to it.
The point was to test out stay NAS vs SAN. I would use a physical machine as the VOB/View server(s) in a final solution. I wasn't expecting this performance hit. If I find the issue, I'll repost. I have an open case with IBM.