Data reported to the WLCG for November, 2009

Page Contents:

• Per-site WLCG Accounting Summary
• USATLAS T2 WLCG Accounting Data
• USCMS T2 WLCG Accounting Data
• GIP Subcluster Summary
• Site Normalization Factors
• Site Availability Measurements
• Available Data Summary

This data was reported at 2009-11-07 01:15:05

(Note: Click on the column headers to sort the table by that column.)

---

WLCG Pledge Information

Each WLCG site has pledged a certain amount of kSI2K for their site; in this section, we compare the amount of pledged KSI2K CPU hours to this month's accounting data. The accounting data is measured in KSI2K-hours, which is the number of hours recorded by the accounting system multiplied by the site normalization factor (shown above). We show data using both CPU hours and Wall (elapsed) hours. Each WLCG site should attempt to hit this month's goal, which is the KSI2K pledge amount times the number of hours in the month times a CPU efficiency factor. The CPU efficiency factor is currently 60%. The table below shows:

• 2008 KSI2K Pledge: 2008 KSI2K pledge in the WLCG MoU
• Month goal of CPU hours: The corresponding number of CPU hours which should be achieved this month (assumes a CPU efficiency of 60% from the pledged amount).
• Normalized CPU hours for all WLCG The number of normalized CPU hours contributed to all WLCG VOs (atlas, cms, alice, lhcb).
• Measured KSI2KThe measured KSI2K size from the GIP.
• Owner VO's KSI2K-Availability Explained below. For each day of the month, multiply that day's available KSI2K times the owner VO's ownership percentage times the RSV-measured availability, then sum them up.
• Normalized wall hours for owner The number of normalized wall hours contributed to the VO which owns the site.
• Normalized wall hours for all WLCG The number of normalized wall hours contributed to all WLCG VOs (atlas, cms, alice, lhcb).
• Normalized wall hours for all VOs The number of normalized wall hours contributed to all VOs.
• Percent of WLCG goal The percentage of the entire months's goal delivered. This is computed by (Normalized CPU hours for WLCG)/(monthly goal).
• Percent of site's time delivered to non-WLCG VOs The percentage of the normalized wall hours delivered to non-WLCG VOs out of the site's total for this month

Please note that the mapping from OSG site to WLCG site is not one-to-one. Sometimes, several OSG sites put together compose a single WLCG site. Further, the WLCG names are not the same as the OSG name. For example, Nebraska is known as "T2_US_Nebraska" to the WLCG, or AGLT2 is "US-AGLT2".

ATLAS Sites

CMS Sites

---

GIP Subcluster Information

In order to generate an accurate normalization factor, the site must advertise the make-up of each cluster using the GIP. Each cluster is divided into heterogeneous (in terms of CPU model) subclusters. When the GIP configuration is run, the site admin specifies, for each subcluster:

• Subcluster name (must be unique)
• CPU Processor model
• Number of cores per subcluster
• (Optional override using alter-attributes.conf) The SI2K score per core.

The normalization factor is the average SI2K score, weighted by number of cores, then divided by 1000.

The process of changing the SI2K score per core through the alter-attributes.conf is documented here

The lookup table for CPU normalization is posted here.

The table below shows the GIP-recorded subcluster information for each site.

This data was reported at 2009-11-06 00:01:03.

---

Availability Data

(WARNING: Experimental metric). A site's KSI2K-availability is calculated by multiplying the following three values, calculated daily, then summed over the whole month

• The total cluster size, measured in KSI2K.
• The ownership percentage of the sponsoring VO.
• The RSV-measured daily availability percentage

Specifically, the first two bullet points are dependent on site admins giving the correct inputs (which is getting more accurate as site admins are being educated in the GIP). The third bullet point is dependent on the reliability if the RSV measurements, which is a new component to the OSG.
The below table shows, for each site, the average monthly value of the above bullet points, in addition to the total KSI2K-availability for the month. The first three values cannot be multiplied together to get the fourth; the first three are averages over the whole month, while the fourth is derived by computing the daily amount.

---

Site Normalization Calculation

The table below shows the calculated normalization constant (using the above table), the normalization constant reported to the WLCG, and the percent difference. Nominally, all WLCG sites should focus on switching to the GIP-based normalization constant from the manually-update one.

---

Data Summary

Finally, the following table shows the data available for each site. Ideally, each site should have an accompanying WLCG Accounting name and more than one GIP subcluster; any incorrect data in the above tables is usually due to one or both of these missing.