[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] Tier 3 hardware requirements

Thanks Brian!

Althought this proyect is just beginning, our goal is to serve ATLAS since Argentina (maybe sharing resources with Chile) and I must study costs and hardware requirements. Now, I am uploading a local HTCondor pool on this institute for minimal related jobs (about 60 cores) but in a near future, we gonna make an investment to reach T3 level.


Ing. Alejandro AcuÃa
IFLP - Conicet
113 63 y 64, La Plata, Bs As, Argentina

El 8/10/19 a las 18:28, Brian Lin escribiÃ:

What experiment will you be serving as a T3? The most accurate hardware
recommendations will come from the experiments themselves as they know
the requirements for their workflows. For example, CMS has Tier-3
suggestions publicly available:
Additionally, are you interested in serving multiple experiments and/or
local users?

That being said, we can provide some pointers for running a generic
HTCondor pool.


For a minimal HTCondor pool, you probably need at least the following
kinds of services:

* A central manager
* Some submit points
* Many execute nodes
* Perhaps some other helpful services

Especially for a smaller site, it may be possible to combine things like
the central manager and one submit point on a single node. Likewise,
other services can often be combined if lightly used.

Rather than any specific hardware suggestions, below are some ratios to
think about when planning hardware for a site:

Central manager:

The minimum requirements here are fairly small. For a cluster of about
10,000 job slots, maybe 1 GB RAM is sufficient.

Submit point:

* Reserve 2 MB RAM for every job slot in the pool. For example, if your
pool has 1000 slots, consider at least 2 GB RAM.

* An additional 3-4 GB RAM per average active user on the submit host.
That is to say, if you have 10 logged-in users on average, you might
want 40 GB RAM. Users often do interactive tasks that may use surprising
amounts of memory.

* 50 GB working storage space (in home or scratch, e.g.) per user on the
submit node. An SSD is preferable here.

Execute nodes:

* At least 2 GB per compute core is a good starting point (and is a
minimum for ATLAS workloads).

* At least 10 GB of disk per job slot.

Other services to consider:
* Squid - a caching proxy for file transfers
* XCache - another file caching system
* perfSONAR - used to measure, record, and report networking performance

- Brian

On 10/3/19 8:52 AM, Alejandro AcuÃa wrote:
Dear Users

I need Tier 3 hardware minimal requirements nowadays. Does anyone know
where I can find them?
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx
with a
subject: Unsubscribe
You can also unsubscribe by visiting

The archives can be found at: