Rates

Overview

The Research Computing Group’s A&FS reviewed recharge rates are authorized to be charged to contracts and grants. Our rates are designed to be easy to incorporate into grant proposals and all of the money recovered from rates is reinvested into HPC resources and salaries. Under certain circumstances, we can also charge equipment funds. Contact us for more information.

Principles of a HPC rate structure

Associates rates:

Should incentive labs to contribute funds to the system(s) in return for access to the system proportional to their anticipated usage.
Should encourage the proper use of the system and discourage misuse.
Should strive for a targeted efficiency for both CPU/MEMORY and storage.
Should be time/resource bound. Does not extend forever.
Should be simple to describe and easy to incorporate into grant submissions.
Should be transparent.

Examples of rates which violate 1, or more, of the above principles.

A cost per user rate, would violate principle #2 (Should encourage the proper use of the system and discourage misuse) as such a rate would encourage account sharing within a lab in order to provide access to a new user without paying associated costs.
A flat rate annual account (for a lab) would violate principle #1 as a flat rate would not be proportional to the labs usage of the system. Heavy users are subsidized by in frequent users.

Associated costs

There are three primary costs associated with maintaining a HPC system.

Cost of hardware (nodes/cpus/memory/disk drives/etc.)
Cost of infrastructure (firewall/switches/head node/controllers/etc.)
Cost of FTE and support (people to maintain the infrastructure, train users, and support of those people [ex. attendance to annual national HPC meeting])

All rates within the core facility are a combination of the above costs, a portion, or all of which, may include subsidize by various entities either because some costs are difficult to recover through the recharge model and/or to encourage a certain type of investment

The CRCF Bioinformatics Core HPC Rate Model

Compute access (dedicated):

Dedicated compute access is supplied through exclusive lab login node. Labs login to the dedicated resource which has access to the submit jobs to the cluster, and not onto the common login node. A lab can control who has access to the dedicated node via the Bioinformatics HPC computing portal. These nodes allow for immediate, or specialized (software) needs. Login nodes are built to spec (5-year warranty); because login nodes aren’t socialized, there is $2000/administration fee/year (100% FTE cost).

It’s possible to later integrate a dedicated node into the shared model for a prorated priority/time and no longer incur the yearly administration fee.

Compute access (shared):

A financial commitment to “system priority” ($1 = 1000 priority points, minimum buy-in is $5K) grants proportional priority access to the whole system. The rate assumes an efficiency(load) of 80% and covers hardware/infrastructure/warranty (~100% of rate) and FTE administration costs are 100% subsidized by ACCD funds to incentivize sharing of resources. Labs should buy in at their approximate expected usage, but can add in additional priority at any time (request via the portal), if they underestimated their usage, or their usage pattern changes. Priority points last for 5 years, the core helps with grant writing to help a lab continue contributing to the system and overall sustainability of the cluster. At the end of 5 years the priority is removed (step function), the lab is not removed from the system, but rather the priority is reduced to 0 (lowest priority) for 1 year, assuming additional priority hasn’t been purchased in the interim.

In the social model, the lab that has contributed the most to the cluster has the highest priority, the lab which contributes the least has the lowest priority. Priority on the system relates to your queue position (among other factors) and is a decay (as you submit jobs) / recover algorithm.

The Core uses the funds to purchase equipment/infrastructure, based on the current usage characteristics of the cluster and is allowed to anticipate the future needs of the user base and purchase accordingly.

The Bioinformatics Core currently has nodes that are 128G RAM - 1TB. Some nodes are equipped with GPU units. Purchasing priority grants access to the whole shared system, a lab’s needs in the future may differ from those at the time of commitment and so does not need to anticipate current and/or future needs down to specific hardware.

The Genome Center supports the Bioinformatics Core's HPC resources by supporting approximately 80% of all the infrastructure costs associated with the system. Genome Center Investigators are then granted access to the shared resources, without needing to "buy-in" (subsidized by the Genome Center) but have a priority value of 0.

Compute access (temporary):

The Bioinformatics Core does not have temporary compute access. However, however we occasionally grant graduate student pilot projects on a case by case basis.

Data Storage:

$100/Tb/Year Primary, not Backed up.
$150/Tb/Year Primary plus back up copy at the University Data Center.
Discounted rate applies to requests >= 100Tb ($167/Tb/3-Years)

Rate assumes an efficiency(load) of 50-60% and covers hardware/infrastructure/warranty (~75% of rate) and FTE administration costs (~25% of rate).

Many labs have both backup up and not backup shares.

To increase to storage is available on request via the Bioinformatics HPC computing portal.

Systems Administration hourly rate:

A rate of $151/hour to cover associated FTE costs for special projects.

HPC Rates

Bioinformatics Core HPC resources are only available to UC Davis Investigators.

Hourly Rate	$151/hr
Cost of Terabyte of Primary Storage	$100/yr
Cost of Terabyte of Backup Storage	$50/yr
Priority on the shared computing system (1000 priority points) - 5 years	$1
Login Node - 5 year warranty	inquire
Administration of a non-shared resource	$2000/year