Cloud.com provides cloud and infrastructure orchestration software for service providers. They have an extensive internal dev/ops environment.
Cloud.com wanted to re-purpose few of their expensive Dell C6220 servers to run additional applications, which meant that they needed to increase the number of existing VMs deployed on each host. As is often the case, there was plenty of CPU, memory, and networking capacity available on each one of the servers, and it was only storage latencies that started to increase disproportionately with higher VM densities.
Cloud.com decided to look for the cheapest possible solution that would improve storage throughput and latencies, which in turn would facilitate the migration of additional VMs to each VMware host.IT Infrastructure
- VMs and Physical Servers - Cloud.com’s IT had four Dell 4-node C6220 servers running Windows Server VMs on VMware 5.5. Before VirtuCache, there were about 80 VMs running MS Exchange, MS Dynamics, and other enterprise applications in this cluster.
- Storage - 18 TB of storage on a gigabit iSCSI Hitachi storage appliance was used by these servers.
- Workload Characteristics - On an average, less than 4 TB of data changed every day and read-write mix varied widely between 40-60 to 80-20 read-write ratio.
- VMware’s Distributed Resource Scheduler (DRS) functionality was configured to be automatic and aggressive, which ensured that workloads were equally distributed at all times across these 4 physical hosts.
|GAVG as measured using ESXTOP||Before VirtuCache||After VirtuCache|
|Read GAVG||35-1500 ms||0.1 – 6 ms|
|Write GAVG||20 – 600 ms||0.1 – 6 ms|
Virtunet’s Write-Back (Read+Write) Caching Competing with Write-Through (Read-Only) Caching at a Large School District
Host side caching software needs to accelerate both reads and writes especially in light of increasing competition from all-flash arrays. Caching writes is important even for workloads that are read intensive. If we were not to accelerate writes, the reads behind the writes on the same thread will not be accelerated, thus slowing reads as well.
Using TPC-C benchmark, we showed 2x improvement in performance versus a read caching software vendor at a large district.
Typically Hadoop workloads are run on bare-metal servers. However since Stanford’s School of Medicine was 100% virtualized, and because their security, monitoring & management tools were integrated within VMware, it was easier for them to deploy Hadoop within VMs, instead of provisioning new physical servers.
The biggest challenge Stanford faced with Hadoop VMs was low throughput and high latencies for writes to disk.
Deploying VirtuCache resulted in write latencies in Hadoop reducing to an eighth of what they were before.
EHR, imaging, analytics, desktop & application virtualization applications are prone to storage performance issues in VMware, that are easily solved with VirtuCache.
VirtuCache Improves the Performance of Equallogic Appliances for VMware VMs and Bare-Metal Linux Servers
By deploying VirtuCache in Dun & Bradstreet's VMware environment, we demonstrated sub 10ms VM latencies for high, write-intensive, random storage IO, which in turn facilitated the P2V of their database servers.
Performance benchmark using Iometer comparing Storage I/O with and without VirtuCache, with multi-threaded 67% read 33% write workload that saturated the disk.
France Telecom is one of the world’s leading telecommunications operators, with a presence in 32 countries, and a customer base of 232 million. France Telecom wanted to provide the lowest dollar per virtual desktop to their subscribers.